projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Linux 4.0-rc7
[deliverable/linux.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
5a0e3ad6 9#include <linux/slab.h>
31b8006e
SW		 10#include <linux/socket.h>
11#include <linux/string.h>
3ebc21f7 12#ifdef CONFIG_BLOCK
68b4476b 13#include <linux/bio.h>
3ebc21f7 14#endif /* CONFIG_BLOCK */
ee3b56f2 15#include <linux/dns_resolver.h>
31b8006e
SW		 16#include <net/tcp.h>
17
2b3e0c90 18#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 19#include <linux/ceph/libceph.h>
20#include <linux/ceph/messenger.h>
21#include <linux/ceph/decode.h>
22#include <linux/ceph/pagelist.h>
bc3b2d7f 23#include <linux/export.h>
31b8006e 24
fe38a2b6
AE		 25#define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
27
31b8006e
SW		 28/*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
bc18f4b1
AE		 37/*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
fbb85a47
SW		 56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
bc18f4b1
AE		 67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
ce2c8903
AE		 78
79#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
8dacc7da
SW		 85/*
86 * connection states
87 */
88#define CON_STATE_CLOSED 1 /* -> PREOPEN */
89#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94
4a861692
SW		 95/*
96 * ceph_connection flag bits
97 */
98#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 104
c9ffc77a
AE		 105static bool con_flag_valid(unsigned long con_flag)
106{
107 switch (con_flag) {
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
113 return true;
114 default:
115 return false;
116 }
117}
118
119static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120{
121 BUG_ON(!con_flag_valid(con_flag));
122
123 clear_bit(con_flag, &con->flags);
124}
125
126static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127{
128 BUG_ON(!con_flag_valid(con_flag));
129
130 set_bit(con_flag, &con->flags);
131}
132
133static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134{
135 BUG_ON(!con_flag_valid(con_flag));
136
137 return test_bit(con_flag, &con->flags);
138}
139
140static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
142{
143 BUG_ON(!con_flag_valid(con_flag));
144
145 return test_and_clear_bit(con_flag, &con->flags);
146}
147
148static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
150{
151 BUG_ON(!con_flag_valid(con_flag));
152
153 return test_and_set_bit(con_flag, &con->flags);
154}
155
e3d5d638
AE		 156/* Slab caches for frequently-allocated structures */
157
158static struct kmem_cache *ceph_msg_cache;
81b36be4 159static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 160
31b8006e
SW		 161/* static tag bytes (protocol control messages) */
162static char tag_msg = CEPH_MSGR_TAG_MSG;
163static char tag_ack = CEPH_MSGR_TAG_ACK;
164static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165
a6a5349d
SW		 166#ifdef CONFIG_LOCKDEP
167static struct lock_class_key socket_class;
168#endif
169
84495f49
AE		 170/*
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
173 */
174#define SKIP_BUF_SIZE 1024
31b8006e
SW		 175
176static void queue_con(struct ceph_connection *con);
37ab77ac 177static void cancel_con(struct ceph_connection *con);
31b8006e 178static void con_work(struct work_struct *);
93209264 179static void con_fault(struct ceph_connection *con);
31b8006e 180
31b8006e 181/*
f64a9317
AE		 182 * Nicely render a sockaddr as a string. An array of formatted
183 * strings is used, to approximate reentrancy.
31b8006e 184 */
f64a9317
AE		 185#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
186#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
187#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
188#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
189
190static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
191static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 192
57666519 193static struct page *zero_page; /* used in certain error cases */
57666519 194
3d14c5d2 195const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 196{
197 int i;
198 char *s;
99f0f3b2
AE		 199 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
200 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 201
f64a9317 202 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 203 s = addr_str[i];
204
205 switch (ss->ss_family) {
206 case AF_INET:
bd406145
AE		 207 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
208 ntohs(in4->sin_port));
31b8006e
SW		 209 break;
210
211 case AF_INET6:
bd406145
AE		 212 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
213 ntohs(in6->sin6_port));
31b8006e
SW		 214 break;
215
216 default:
d3002b97
AE		 217 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
218 ss->ss_family);
31b8006e
SW		 219 }
220
221 return s;
222}
3d14c5d2 223EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 224
63f2d211
SW		 225static void encode_my_addr(struct ceph_messenger *msgr)
226{
227 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
228 ceph_encode_addr(&msgr->my_enc_addr);
229}
230
31b8006e
SW		 231/*
232 * work queue for all reading and writing to/from the socket.
233 */
e0f43c94 234static struct workqueue_struct *ceph_msgr_wq;
31b8006e 235
e3d5d638
AE		 236static int ceph_msgr_slab_init(void)
237{
238 BUG_ON(ceph_msg_cache);
239 ceph_msg_cache = kmem_cache_create("ceph_msg",
240 sizeof (struct ceph_msg),
241 __alignof__(struct ceph_msg), 0, NULL);
81b36be4
AE		 242
243 if (!ceph_msg_cache)
244 return -ENOMEM;
245
246 BUG_ON(ceph_msg_data_cache);
247 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
248 sizeof (struct ceph_msg_data),
249 __alignof__(struct ceph_msg_data),
250 0, NULL);
251 if (ceph_msg_data_cache)
252 return 0;
253
254 kmem_cache_destroy(ceph_msg_cache);
255 ceph_msg_cache = NULL;
256
257 return -ENOMEM;
e3d5d638
AE		 258}
259
260static void ceph_msgr_slab_exit(void)
261{
81b36be4
AE		 262 BUG_ON(!ceph_msg_data_cache);
263 kmem_cache_destroy(ceph_msg_data_cache);
264 ceph_msg_data_cache = NULL;
265
e3d5d638
AE		 266 BUG_ON(!ceph_msg_cache);
267 kmem_cache_destroy(ceph_msg_cache);
268 ceph_msg_cache = NULL;
269}
270
15417167 271static void _ceph_msgr_exit(void)
6173d1f0 272{
d3002b97 273 if (ceph_msgr_wq) {
6173d1f0 274 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 275 ceph_msgr_wq = NULL;
276 }
6173d1f0 277
e3d5d638
AE		 278 ceph_msgr_slab_exit();
279
6173d1f0
AE		 280 BUG_ON(zero_page == NULL);
281 kunmap(zero_page);
282 page_cache_release(zero_page);
283 zero_page = NULL;
284}
285
3d14c5d2 286int ceph_msgr_init(void)
31b8006e 287{
57666519
AE		 288 BUG_ON(zero_page != NULL);
289 zero_page = ZERO_PAGE(0);
290 page_cache_get(zero_page);
291
e3d5d638
AE		 292 if (ceph_msgr_slab_init())
293 return -ENOMEM;
294
f9865f06
ID		 295 /*
296 * The number of active work items is limited by the number of
297 * connections, so leave @max_active at default.
298 */
299 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 300 if (ceph_msgr_wq)
301 return 0;
57666519 302
6173d1f0
AE		 303 pr_err("msgr_init failed to create workqueue\n");
304 _ceph_msgr_exit();
57666519 305
6173d1f0 306 return -ENOMEM;
31b8006e 307}
3d14c5d2 308EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 309
310void ceph_msgr_exit(void)
311{
57666519 312 BUG_ON(ceph_msgr_wq == NULL);
57666519 313
6173d1f0 314 _ceph_msgr_exit();
31b8006e 315}
3d14c5d2 316EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 317
cd84db6e 318void ceph_msgr_flush(void)
a922d38f
SW		 319{
320 flush_workqueue(ceph_msgr_wq);
321}
3d14c5d2 322EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 323
ce2c8903
AE		 324/* Connection socket state transition functions */
325
326static void con_sock_state_init(struct ceph_connection *con)
327{
328 int old_state;
329
330 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
331 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
332 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 333 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
334 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 335}
336
337static void con_sock_state_connecting(struct ceph_connection *con)
338{
339 int old_state;
340
341 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
342 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
343 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 344 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
345 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 346}
347
348static void con_sock_state_connected(struct ceph_connection *con)
349{
350 int old_state;
351
352 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
353 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
354 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 355 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
356 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 357}
358
359static void con_sock_state_closing(struct ceph_connection *con)
360{
361 int old_state;
362
363 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
364 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
365 old_state != CON_SOCK_STATE_CONNECTED &&
366 old_state != CON_SOCK_STATE_CLOSING))
367 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 368 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
369 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 370}
371
372static void con_sock_state_closed(struct ceph_connection *con)
373{
374 int old_state;
375
376 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
377 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 378 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 379 old_state != CON_SOCK_STATE_CONNECTING &&
380 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 381 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 382 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
383 CON_SOCK_STATE_CLOSED);
ce2c8903 384}
a922d38f 385
31b8006e
SW		 386/*
387 * socket callback functions
388 */
389
390/* data available on socket, or listen socket received a connect */
676d2369 391static void ceph_sock_data_ready(struct sock *sk)
31b8006e 392{
bd406145 393 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 394 if (atomic_read(&con->msgr->stopping)) {
395 return;
396 }
bd406145 397
31b8006e 398 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 399 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 400 con, con->state);
401 queue_con(con);
402 }
403}
404
405/* socket has buffer space for writing */
327800bd 406static void ceph_sock_write_space(struct sock *sk)
31b8006e 407{
d3002b97 408 struct ceph_connection *con = sk->sk_user_data;
31b8006e 409
182fac26
JS		 410 /* only queue to workqueue if there is data we want to write,
411 * and there is sufficient space in the socket buffer to accept
327800bd 412 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 413 * doesn't get called again until try_write() fills the socket
414 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
415 * and net/core/stream.c:sk_stream_write_space().
416 */
c9ffc77a 417 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 418 if (sk_stream_is_writeable(sk)) {
327800bd 419 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 420 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
421 queue_con(con);
422 }
31b8006e 423 } else {
327800bd 424 dout("%s %p nothing to write\n", __func__, con);
31b8006e 425 }
31b8006e
SW		 426}
427
428/* socket's state has changed */
327800bd 429static void ceph_sock_state_change(struct sock *sk)
31b8006e 430{
bd406145 431 struct ceph_connection *con = sk->sk_user_data;
31b8006e 432
327800bd 433 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 434 con, con->state, sk->sk_state);
435
31b8006e
SW		 436 switch (sk->sk_state) {
437 case TCP_CLOSE:
327800bd 438 dout("%s TCP_CLOSE\n", __func__);
31b8006e 439 case TCP_CLOSE_WAIT:
327800bd 440 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 441 con_sock_state_closing(con);
c9ffc77a 442 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 443 queue_con(con);
31b8006e
SW		 444 break;
445 case TCP_ESTABLISHED:
327800bd 446 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 447 con_sock_state_connected(con);
31b8006e
SW		 448 queue_con(con);
449 break;
d3002b97
AE		 450 default: /* Everything else is uninteresting */
451 break;
31b8006e
SW		 452 }
453}
454
455/*
456 * set up socket callbacks
457 */
458static void set_sock_callbacks(struct socket *sock,
459 struct ceph_connection *con)
460{
461 struct sock *sk = sock->sk;
bd406145 462 sk->sk_user_data = con;
327800bd
AE		 463 sk->sk_data_ready = ceph_sock_data_ready;
464 sk->sk_write_space = ceph_sock_write_space;
465 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 466}
467
468
469/*
470 * socket helpers
471 */
472
473/*
474 * initiate connection to a remote socket.
475 */
41617d0c 476static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 477{
f91d3471 478 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 479 struct socket *sock;
480 int ret;
481
482 BUG_ON(con->sock);
f91d3471
SW		 483 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
484 IPPROTO_TCP, &sock);
31b8006e 485 if (ret)
41617d0c 486 return ret;
89baaa57 487 sock->sk->sk_allocation = GFP_NOFS | __GFP_MEMALLOC;
31b8006e 488
a6a5349d
SW		 489#ifdef CONFIG_LOCKDEP
490 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
491#endif
492
31b8006e
SW		 493 set_sock_callbacks(sock, con);
494
3d14c5d2 495 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 496
89a86be0 497 con_sock_state_connecting(con);
f91d3471
SW		 498 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 O_NONBLOCK);
31b8006e
SW		 500 if (ret == -EINPROGRESS) {
501 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 502 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 503 sock->sk->sk_state);
a5bc3129 504 } else if (ret < 0) {
31b8006e 505 pr_err("connect %s error %d\n",
3d14c5d2 506 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 507 sock_release(sock);
31b8006e 508 con->error_msg = "connect error";
31b8006e 509
41617d0c 510 return ret;
a5bc3129 511 }
89baaa57 512
ba988f87
CH		 513 if (con->msgr->tcp_nodelay) {
514 int optval = 1;
515
516 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
517 (char *)&optval, sizeof(optval));
518 if (ret)
519 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
520 ret);
521 }
522
89baaa57
MC		 523 sk_set_memalloc(sock->sk);
524
a5bc3129 525 con->sock = sock;
41617d0c 526 return 0;
31b8006e
SW		 527}
528
529static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
530{
531 struct kvec iov = {buf, len};
532 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 533 int r;
31b8006e 534
98bdb0aa
SW		 535 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
536 if (r == -EAGAIN)
537 r = 0;
538 return r;
31b8006e
SW		 539}
540
afb3d90e
AE		 541static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
542 int page_offset, size_t length)
543{
544 void *kaddr;
545 int ret;
546
547 BUG_ON(page_offset + length > PAGE_SIZE);
548
549 kaddr = kmap(page);
550 BUG_ON(!kaddr);
551 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
552 kunmap(page);
553
554 return ret;
555}
556
31b8006e
SW		 557/*
558 * write something. @more is true if caller will be sending more data
559 * shortly.
560 */
561static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
562 size_t kvlen, size_t len, int more)
563{
564 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 565 int r;
31b8006e
SW		 566
567 if (more)
568 msg.msg_flags |= MSG_MORE;
569 else
570 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
571
42961d23
SW		 572 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
573 if (r == -EAGAIN)
574 r = 0;
575 return r;
31b8006e
SW		 576}
577
178eda29 578static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 579 int offset, size_t size, bool more)
31739139
AE		 580{
581 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
582 int ret;
583
584 ret = kernel_sendpage(sock, page, offset, size, flags);
585 if (ret == -EAGAIN)
586 ret = 0;
587
588 return ret;
589}
590
178eda29
CC		 591static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
592 int offset, size_t size, bool more)
593{
594 int ret;
595 struct kvec iov;
596
597 /* sendpage cannot properly handle pages with page_count == 0,
598 * we need to fallback to sendmsg if that's the case */
599 if (page_count(page) >= 1)
600 return __ceph_tcp_sendpage(sock, page, offset, size, more);
601
602 iov.iov_base = kmap(page) + offset;
603 iov.iov_len = size;
604 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
605 kunmap(page);
606
607 return ret;
608}
31b8006e
SW		 609
610/*
611 * Shutdown/close the socket for the given connection.
612 */
613static int con_close_socket(struct ceph_connection *con)
614{
8007b8d6 615 int rc = 0;
31b8006e
SW		 616
617 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 618 if (con->sock) {
619 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
620 sock_release(con->sock);
621 con->sock = NULL;
622 }
456ea468
AE		 623
624 /*
4a861692 625 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 626 * independent of the connection mutex, and we could have
627 * received a socket close event before we had the chance to
628 * shut the socket down.
629 */
c9ffc77a 630 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 631
ce2c8903 632 con_sock_state_closed(con);
31b8006e
SW		 633 return rc;
634}
635
636/*
637 * Reset a connection. Discard all incoming and outgoing messages
638 * and clear *_seq state.
639 */
640static void ceph_msg_remove(struct ceph_msg *msg)
641{
642 list_del_init(&msg->list_head);
38941f80 643 BUG_ON(msg->con == NULL);
36eb71aa 644 msg->con->ops->put(msg->con);
38941f80
AE		 645 msg->con = NULL;
646
31b8006e
SW		 647 ceph_msg_put(msg);
648}
649static void ceph_msg_remove_list(struct list_head *head)
650{
651 while (!list_empty(head)) {
652 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
653 list_head);
654 ceph_msg_remove(msg);
655 }
656}
657
658static void reset_connection(struct ceph_connection *con)
659{
660 /* reset connection, out_queue, msg_ and connect_seq */
661 /* discard existing out_queue and msg_seq */
0fa6ebc6 662 dout("reset_connection %p\n", con);
31b8006e
SW		 663 ceph_msg_remove_list(&con->out_queue);
664 ceph_msg_remove_list(&con->out_sent);
665
cf3e5c40 666 if (con->in_msg) {
38941f80
AE		 667 BUG_ON(con->in_msg->con != con);
668 con->in_msg->con = NULL;
cf3e5c40
SW		 669 ceph_msg_put(con->in_msg);
670 con->in_msg = NULL;
36eb71aa 671 con->ops->put(con);
cf3e5c40
SW		 672 }
673
31b8006e
SW		 674 con->connect_seq = 0;
675 con->out_seq = 0;
c86a2930
SW		 676 if (con->out_msg) {
677 ceph_msg_put(con->out_msg);
678 con->out_msg = NULL;
679 }
31b8006e 680 con->in_seq = 0;
0e0d5e0c 681 con->in_seq_acked = 0;
31b8006e
SW		 682}
683
684/*
685 * mark a peer down. drop any open connections.
686 */
687void ceph_con_close(struct ceph_connection *con)
688{
8c50c817 689 mutex_lock(&con->mutex);
3d14c5d2
YS		 690 dout("con_close %p peer %s\n", con,
691 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 692 con->state = CON_STATE_CLOSED;
a5988c49 693
c9ffc77a
AE		 694 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
695 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
696 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
697 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 698
31b8006e 699 reset_connection(con);
6f2bc3ff 700 con->peer_global_seq = 0;
37ab77ac 701 cancel_con(con);
ee76e073 702 con_close_socket(con);
ec302645 703 mutex_unlock(&con->mutex);
31b8006e 704}
3d14c5d2 705EXPORT_SYMBOL(ceph_con_close);
31b8006e 706
31b8006e
SW		 707/*
708 * Reopen a closed connection, with a new peer address.
709 */
b7a9e5dd
SW		 710void ceph_con_open(struct ceph_connection *con,
711 __u8 entity_type, __u64 entity_num,
712 struct ceph_entity_addr *addr)
31b8006e 713{
5469155f 714 mutex_lock(&con->mutex);
3d14c5d2 715 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 716
122070a2 717 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 718 con->state = CON_STATE_PREOPEN;
a5988c49 719
b7a9e5dd
SW		 720 con->peer_name.type = (__u8) entity_type;
721 con->peer_name.num = cpu_to_le64(entity_num);
722
31b8006e 723 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 724 con->delay = 0; /* reset backoff memory */
5469155f 725 mutex_unlock(&con->mutex);
31b8006e
SW		 726 queue_con(con);
727}
3d14c5d2 728EXPORT_SYMBOL(ceph_con_open);
31b8006e 729
87b315a5
SW		 730/*
731 * return true if this connection ever successfully opened
732 */
733bool ceph_con_opened(struct ceph_connection *con)
734{
735 return con->connect_seq > 0;
736}
737
31b8006e
SW		 738/*
739 * initialize a new connection.
740 */
1bfd89f4
AE		 741void ceph_con_init(struct ceph_connection *con, void *private,
742 const struct ceph_connection_operations *ops,
b7a9e5dd 743 struct ceph_messenger *msgr)
31b8006e
SW		 744{
745 dout("con_init %p\n", con);
746 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 747 con->private = private;
748 con->ops = ops;
31b8006e 749 con->msgr = msgr;
ce2c8903
AE		 750
751 con_sock_state_init(con);
752
ec302645 753 mutex_init(&con->mutex);
31b8006e
SW		 754 INIT_LIST_HEAD(&con->out_queue);
755 INIT_LIST_HEAD(&con->out_sent);
756 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 757
8dacc7da 758 con->state = CON_STATE_CLOSED;
31b8006e 759}
3d14c5d2 760EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 761
762
763/*
764 * We maintain a global counter to order connection attempts. Get
765 * a unique seq greater than @gt.
766 */
767static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
768{
769 u32 ret;
770
771 spin_lock(&msgr->global_seq_lock);
772 if (msgr->global_seq < gt)
773 msgr->global_seq = gt;
774 ret = ++msgr->global_seq;
775 spin_unlock(&msgr->global_seq_lock);
776 return ret;
777}
778
e2200423 779static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 780{
781 con->out_kvec_left = 0;
782 con->out_kvec_bytes = 0;
783 con->out_kvec_cur = &con->out_kvec[0];
784}
785
e2200423 786static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 787 size_t size, void *data)
788{
789 int index;
790
791 index = con->out_kvec_left;
792 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
793
794 con->out_kvec[index].iov_len = size;
795 con->out_kvec[index].iov_base = data;
796 con->out_kvec_left++;
797 con->out_kvec_bytes += size;
798}
31b8006e 799
df6ad1f9 800#ifdef CONFIG_BLOCK
6aaa4511
AE		 801
802/*
803 * For a bio data item, a piece is whatever remains of the next
804 * entry in the current bio iovec, or the first entry in the next
805 * bio in the list.
806 */
8ae4f4f5 807static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 808 size_t length)
6aaa4511 809{
8ae4f4f5 810 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 811 struct bio *bio;
812
813 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
814
815 bio = data->bio;
816 BUG_ON(!bio);
6aaa4511 817
ca8b3a69 818 cursor->resid = min(length, data->bio_length);
6aaa4511 819 cursor->bio = bio;
f38a5181
KO		 820 cursor->bvec_iter = bio->bi_iter;
821 cursor->last_piece =
822 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 823}
824
8ae4f4f5 825static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 826 size_t *page_offset,
827 size_t *length)
828{
8ae4f4f5 829 struct ceph_msg_data *data = cursor->data;
6aaa4511 830 struct bio *bio;
f38a5181 831 struct bio_vec bio_vec;
6aaa4511
AE		 832
833 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
834
835 bio = cursor->bio;
836 BUG_ON(!bio);
837
f38a5181 838 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 839
f38a5181 840 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 841 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 842 if (cursor->last_piece) /* pagelist offset is always 0 */
843 *length = cursor->resid;
844 else
f38a5181 845 *length = (size_t) bio_vec.bv_len;
25aff7c5 846 BUG_ON(*length > cursor->resid);
5df521b1 847 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 848
f38a5181 849 return bio_vec.bv_page;
6aaa4511
AE		 850}
851
8ae4f4f5
AE		 852static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
853 size_t bytes)
6aaa4511 854{
6aaa4511 855 struct bio *bio;
f38a5181 856 struct bio_vec bio_vec;
6aaa4511 857
8ae4f4f5 858 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 859
860 bio = cursor->bio;
861 BUG_ON(!bio);
862
f38a5181 863 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 864
865 /* Advance the cursor offset */
866
25aff7c5
AE		 867 BUG_ON(cursor->resid < bytes);
868 cursor->resid -= bytes;
f38a5181
KO		 869
870 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
871
872 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 873 return false; /* more bytes to process in this segment */
874
875 /* Move on to the next segment, and possibly the next bio */
876
f38a5181 877 if (!cursor->bvec_iter.bi_size) {
6aaa4511 878 bio = bio->bi_next;
0ec1d15e
ID		 879 cursor->bio = bio;
880 if (bio)
881 cursor->bvec_iter = bio->bi_iter;
882 else
883 memset(&cursor->bvec_iter, 0,
884 sizeof(cursor->bvec_iter));
6aaa4511 885 }
6aaa4511 886
25aff7c5
AE		 887 if (!cursor->last_piece) {
888 BUG_ON(!cursor->resid);
889 BUG_ON(!bio);
890 /* A short read is OK, so use <= rather than == */
f38a5181 891 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 892 cursor->last_piece = true;
25aff7c5 893 }
6aaa4511
AE		 894
895 return true;
896}
ea96571f 897#endif /* CONFIG_BLOCK */
df6ad1f9 898
e766d7b5
AE		 899/*
900 * For a page array, a piece comes from the first page in the array
901 * that has not already been fully consumed.
902 */
8ae4f4f5 903static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 904 size_t length)
e766d7b5 905{
8ae4f4f5 906 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 907 int page_count;
908
909 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
910
911 BUG_ON(!data->pages);
912 BUG_ON(!data->length);
913
ca8b3a69 914 cursor->resid = min(length, data->length);
e766d7b5 915 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 916 cursor->page_offset = data->alignment & ~PAGE_MASK;
917 cursor->page_index = 0;
56fc5659
AE		 918 BUG_ON(page_count > (int)USHRT_MAX);
919 cursor->page_count = (unsigned short)page_count;
920 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 921 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 922}
923
8ae4f4f5
AE		 924static struct page *
925ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
926 size_t *page_offset, size_t *length)
e766d7b5 927{
8ae4f4f5 928 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 929
930 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
931
932 BUG_ON(cursor->page_index >= cursor->page_count);
933 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 934
935 *page_offset = cursor->page_offset;
25aff7c5 936 if (cursor->last_piece)
e766d7b5 937 *length = cursor->resid;
25aff7c5 938 else
e766d7b5 939 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 940
941 return data->pages[cursor->page_index];
942}
943
8ae4f4f5 944static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 945 size_t bytes)
946{
8ae4f4f5 947 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 948
949 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 950
951 /* Advance the cursor page offset */
952
953 cursor->resid -= bytes;
5df521b1
AE		 954 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
955 if (!bytes || cursor->page_offset)
e766d7b5
AE		 956 return false; /* more bytes to process in the current page */
957
d90deda6
YZ		 958 if (!cursor->resid)
959 return false; /* no more data */
960
5df521b1 961 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 962
963 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 964 cursor->page_index++;
25aff7c5 965 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 966
967 return true;
968}
969
fe38a2b6 970/*
dd236fcb
AE		 971 * For a pagelist, a piece is whatever remains to be consumed in the
972 * first page in the list, or the front of the next page.
fe38a2b6 973 */
8ae4f4f5
AE		 974static void
975ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 976 size_t length)
fe38a2b6 977{
8ae4f4f5 978 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 979 struct ceph_pagelist *pagelist;
980 struct page *page;
981
dd236fcb 982 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 983
984 pagelist = data->pagelist;
985 BUG_ON(!pagelist);
25aff7c5
AE		 986
987 if (!length)
fe38a2b6
AE		 988 return; /* pagelist can be assigned but empty */
989
990 BUG_ON(list_empty(&pagelist->head));
991 page = list_first_entry(&pagelist->head, struct page, lru);
992
ca8b3a69 993 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 994 cursor->page = page;
995 cursor->offset = 0;
a51b272e 996 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 997}
998
8ae4f4f5
AE		 999static struct page *
1000ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1001 size_t *page_offset, size_t *length)
fe38a2b6 1002{
8ae4f4f5 1003 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1004 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1005
1006 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1007
1008 pagelist = data->pagelist;
1009 BUG_ON(!pagelist);
1010
1011 BUG_ON(!cursor->page);
25aff7c5 1012 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1013
5df521b1 1014 /* offset of first page in pagelist is always 0 */
fe38a2b6 1015 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1016 if (cursor->last_piece)
25aff7c5
AE		 1017 *length = cursor->resid;
1018 else
1019 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1020
8ae4f4f5 1021 return cursor->page;
fe38a2b6
AE		 1022}
1023
8ae4f4f5 1024static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1025 size_t bytes)
fe38a2b6 1026{
8ae4f4f5 1027 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1028 struct ceph_pagelist *pagelist;
1029
1030 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1031
1032 pagelist = data->pagelist;
1033 BUG_ON(!pagelist);
25aff7c5
AE		 1034
1035 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1036 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1037
1038 /* Advance the cursor offset */
1039
25aff7c5 1040 cursor->resid -= bytes;
fe38a2b6 1041 cursor->offset += bytes;
5df521b1 1042 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1043 if (!bytes || cursor->offset & ~PAGE_MASK)
1044 return false; /* more bytes to process in the current page */
1045
d90deda6
YZ		 1046 if (!cursor->resid)
1047 return false; /* no more data */
1048
fe38a2b6
AE		 1049 /* Move on to the next page */
1050
1051 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1052 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 1053 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1054
1055 return true;
1056}
1057
dd236fcb
AE		 1058/*
1059 * Message data is handled (sent or received) in pieces, where each
1060 * piece resides on a single page. The network layer might not
1061 * consume an entire piece at once. A data item's cursor keeps
1062 * track of which piece is next to process and how much remains to
1063 * be processed in that piece. It also tracks whether the current
1064 * piece is the last one in the data item.
1065 */
ca8b3a69 1066static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1067{
ca8b3a69 1068 size_t length = cursor->total_resid;
8ae4f4f5 1069
8ae4f4f5 1070 switch (cursor->data->type) {
dd236fcb 1071 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1072 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1073 break;
e766d7b5 1074 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1075 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1076 break;
dd236fcb
AE		 1077#ifdef CONFIG_BLOCK
1078 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1079 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1080 break;
dd236fcb 1081#endif /* CONFIG_BLOCK */
6aaa4511 1082 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1083 default:
1084 /* BUG(); */
1085 break;
1086 }
8ae4f4f5 1087 cursor->need_crc = true;
dd236fcb
AE		 1088}
1089
ca8b3a69
AE		 1090static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1091{
1092 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1093 struct ceph_msg_data *data;
1094
1095 BUG_ON(!length);
1096 BUG_ON(length > msg->data_length);
1097 BUG_ON(list_empty(&msg->data));
1098
ca8b3a69
AE		 1099 cursor->data_head = &msg->data;
1100 cursor->total_resid = length;
1101 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1102 cursor->data = data;
1103
1104 __ceph_msg_data_cursor_init(cursor);
1105}
1106
dd236fcb
AE		 1107/*
1108 * Return the page containing the next piece to process for a given
1109 * data item, and supply the page offset and length of that piece.
1110 * Indicate whether this is the last piece in this data item.
1111 */
8ae4f4f5
AE		 1112static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1113 size_t *page_offset, size_t *length,
dd236fcb
AE		 1114 bool *last_piece)
1115{
1116 struct page *page;
1117
8ae4f4f5 1118 switch (cursor->data->type) {
dd236fcb 1119 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1120 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1121 break;
e766d7b5 1122 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1123 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1124 break;
dd236fcb
AE		 1125#ifdef CONFIG_BLOCK
1126 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1127 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1128 break;
dd236fcb 1129#endif /* CONFIG_BLOCK */
6aaa4511 1130 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1131 default:
1132 page = NULL;
1133 break;
1134 }
1135 BUG_ON(!page);
1136 BUG_ON(*page_offset + *length > PAGE_SIZE);
1137 BUG_ON(!*length);
1138 if (last_piece)
8ae4f4f5 1139 *last_piece = cursor->last_piece;
dd236fcb
AE		 1140
1141 return page;
1142}
1143
1144/*
1145 * Returns true if the result moves the cursor on to the next piece
1146 * of the data item.
1147 */
8ae4f4f5
AE		 1148static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1149 size_t bytes)
dd236fcb
AE		 1150{
1151 bool new_piece;
1152
25aff7c5 1153 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1154 switch (cursor->data->type) {
dd236fcb 1155 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1156 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1157 break;
e766d7b5 1158 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1159 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1160 break;
dd236fcb
AE		 1161#ifdef CONFIG_BLOCK
1162 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1163 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1164 break;
dd236fcb 1165#endif /* CONFIG_BLOCK */
6aaa4511 1166 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1167 default:
1168 BUG();
1169 break;
1170 }
ca8b3a69 1171 cursor->total_resid -= bytes;
dd236fcb 1172
ca8b3a69
AE		 1173 if (!cursor->resid && cursor->total_resid) {
1174 WARN_ON(!cursor->last_piece);
1175 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1176 cursor->data = list_entry_next(cursor->data, links);
1177 __ceph_msg_data_cursor_init(cursor);
a51b272e 1178 new_piece = true;
ca8b3a69 1179 }
a51b272e 1180 cursor->need_crc = new_piece;
ca8b3a69 1181
dd236fcb
AE		 1182 return new_piece;
1183}
1184
98fa5dd8 1185static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1186{
739c905b 1187 BUG_ON(!msg);
25aff7c5 1188 BUG_ON(!data_len);
739c905b 1189
4c59b4a2 1190 /* Initialize data cursor */
fe38a2b6 1191
8ae4f4f5 1192 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1193}
1194
31b8006e
SW		 1195/*
1196 * Prepare footer for currently outgoing message, and finish things
1197 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1198 */
859eb799 1199static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1200{
1201 struct ceph_msg *m = con->out_msg;
859eb799 1202 int v = con->out_kvec_left;
31b8006e 1203
fd154f3c
AE		 1204 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1205
31b8006e
SW		 1206 dout("prepare_write_message_footer %p\n", con);
1207 con->out_kvec_is_msg = true;
1208 con->out_kvec[v].iov_base = &m->footer;
33d07337
YZ		 1209 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1210 if (con->ops->sign_message)
1211 con->ops->sign_message(con, m);
1212 else
1213 m->footer.sig = 0;
1214 con->out_kvec[v].iov_len = sizeof(m->footer);
1215 con->out_kvec_bytes += sizeof(m->footer);
1216 } else {
1217 m->old_footer.flags = m->footer.flags;
1218 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1219 con->out_kvec_bytes += sizeof(m->old_footer);
1220 }
31b8006e
SW		 1221 con->out_kvec_left++;
1222 con->out_more = m->more_to_follow;
c86a2930 1223 con->out_msg_done = true;
31b8006e
SW		 1224}
1225
1226/*
1227 * Prepare headers for the next outgoing message.
1228 */
1229static void prepare_write_message(struct ceph_connection *con)
1230{
1231 struct ceph_msg *m;
a9a0c51a 1232 u32 crc;
31b8006e 1233
e2200423 1234 con_out_kvec_reset(con);
31b8006e 1235 con->out_kvec_is_msg = true;
c86a2930 1236 con->out_msg_done = false;
31b8006e
SW		 1237
1238 /* Sneak an ack in there first? If we can get it into the same
1239 * TCP packet that's a good thing. */
1240 if (con->in_seq > con->in_seq_acked) {
1241 con->in_seq_acked = con->in_seq;
e2200423 1242 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1243 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1244 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1245 &con->out_temp_ack);
31b8006e
SW		 1246 }
1247
38941f80 1248 BUG_ON(list_empty(&con->out_queue));
859eb799 1249 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1250 con->out_msg = m;
38941f80 1251 BUG_ON(m->con != con);
4cf9d544
SW		 1252
1253 /* put message on sent list */
1254 ceph_msg_get(m);
1255 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1256
e84346b7
SW		 1257 /*
1258 * only assign outgoing seq # if we haven't sent this message
1259 * yet. if it is requeued, resend with it's original seq.
1260 */
1261 if (m->needs_out_seq) {
1262 m->hdr.seq = cpu_to_le64(++con->out_seq);
1263 m->needs_out_seq = false;
1264 }
98fa5dd8 1265 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1266
98fa5dd8 1267 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1268 m, con->out_seq, le16_to_cpu(m->hdr.type),
1269 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1270 m->data_length);
31b8006e
SW		 1271 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1272
1273 /* tag + hdr + front + middle */
e2200423
AE		 1274 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1275 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1276 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1277
31b8006e 1278 if (m->middle)
e2200423 1279 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1280 m->middle->vec.iov_base);
31b8006e
SW		 1281
1282 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1283 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1284 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1285 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1286
1287 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1288 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1289 if (m->middle) {
1290 crc = crc32c(0, m->middle->vec.iov_base,
1291 m->middle->vec.iov_len);
1292 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1293 } else
31b8006e 1294 con->out_msg->footer.middle_crc = 0;
739c905b 1295 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1296 le32_to_cpu(con->out_msg->footer.front_crc),
1297 le32_to_cpu(con->out_msg->footer.middle_crc));
1298
1299 /* is there a data payload? */
739c905b 1300 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1301 if (m->data_length) {
1302 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1303 con->out_more = 1; /* data + footer will follow */
1304 } else {
31b8006e 1305 /* no, queue up footer too and be done */
859eb799 1306 prepare_write_message_footer(con);
78625051 1307 }
31b8006e 1308
c9ffc77a 1309 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1310}
1311
1312/*
1313 * Prepare an ack.
1314 */
1315static void prepare_write_ack(struct ceph_connection *con)
1316{
1317 dout("prepare_write_ack %p %llu -> %llu\n", con,
1318 con->in_seq_acked, con->in_seq);
1319 con->in_seq_acked = con->in_seq;
1320
e2200423 1321 con_out_kvec_reset(con);
859eb799 1322
e2200423 1323 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1324
31b8006e 1325 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1326 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1327 &con->out_temp_ack);
1328
31b8006e 1329 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1330 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1331}
1332
3a23083b
SW		 1333/*
1334 * Prepare to share the seq during handshake
1335 */
1336static void prepare_write_seq(struct ceph_connection *con)
1337{
1338 dout("prepare_write_seq %p %llu -> %llu\n", con,
1339 con->in_seq_acked, con->in_seq);
1340 con->in_seq_acked = con->in_seq;
1341
1342 con_out_kvec_reset(con);
1343
1344 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1345 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1346 &con->out_temp_ack);
1347
1348 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1349}
1350
31b8006e
SW		 1351/*
1352 * Prepare to write keepalive byte.
1353 */
1354static void prepare_write_keepalive(struct ceph_connection *con)
1355{
1356 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1357 con_out_kvec_reset(con);
1358 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1359 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1360}
1361
1362/*
1363 * Connection negotiation.
1364 */
1365
dac1e716
AE		 1366static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1367 int *auth_proto)
4e7a5dcd 1368{
a3530df3 1369 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1370
1371 if (!con->ops->get_authorizer) {
1372 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1373 con->out_connect.authorizer_len = 0;
729796be 1374 return NULL;
b1c6b980
AE		 1375 }
1376
1377 /* Can't hold the mutex while getting authorizer */
ec302645 1378 mutex_unlock(&con->mutex);
dac1e716 1379 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1380 mutex_lock(&con->mutex);
4e7a5dcd 1381
a3530df3 1382 if (IS_ERR(auth))
729796be 1383 return auth;
8dacc7da 1384 if (con->state != CON_STATE_NEGOTIATING)
729796be 1385 return ERR_PTR(-EAGAIN);
0da5d703 1386
8f43fb53
AE		 1387 con->auth_reply_buf = auth->authorizer_reply_buf;
1388 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1389 return auth;
4e7a5dcd
SW		 1390}
1391
31b8006e
SW		 1392/*
1393 * We connected to a peer and are saying hello.
1394 */
e825a66d 1395static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1396{
e2200423
AE		 1397 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1398 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1399 &con->msgr->my_enc_addr);
eed0ef2c 1400
eed0ef2c 1401 con->out_more = 0;
c9ffc77a 1402 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1403}
1404
e825a66d 1405static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1406{
95c96174 1407 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1408 int proto;
dac1e716 1409 int auth_proto;
729796be 1410 struct ceph_auth_handshake *auth;
31b8006e
SW		 1411
1412 switch (con->peer_name.type) {
1413 case CEPH_ENTITY_TYPE_MON:
1414 proto = CEPH_MONC_PROTOCOL;
1415 break;
1416 case CEPH_ENTITY_TYPE_OSD:
1417 proto = CEPH_OSDC_PROTOCOL;
1418 break;
1419 case CEPH_ENTITY_TYPE_MDS:
1420 proto = CEPH_MDSC_PROTOCOL;
1421 break;
1422 default:
1423 BUG();
1424 }
1425
1426 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1427 con->connect_seq, global_seq, proto);
4e7a5dcd 1428
e825a66d 1429 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1430 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1431 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1432 con->out_connect.global_seq = cpu_to_le32(global_seq);
1433 con->out_connect.protocol_version = cpu_to_le32(proto);
1434 con->out_connect.flags = 0;
31b8006e 1435
dac1e716
AE		 1436 auth_proto = CEPH_AUTH_UNKNOWN;
1437 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1438 if (IS_ERR(auth))
1439 return PTR_ERR(auth);
3da54776 1440
dac1e716 1441 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1442 con->out_connect.authorizer_len = auth ?
1443 cpu_to_le32(auth->authorizer_buf_len) : 0;
1444
e2200423 1445 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1446 &con->out_connect);
1447 if (auth && auth->authorizer_buf_len)
e2200423 1448 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1449 auth->authorizer_buf);
859eb799 1450
31b8006e 1451 con->out_more = 0;
c9ffc77a 1452 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1453
e10c758e 1454 return 0;
31b8006e
SW		 1455}
1456
31b8006e
SW		 1457/*
1458 * write as much of pending kvecs to the socket as we can.
1459 * 1 -> done
1460 * 0 -> socket full, but more to do
1461 * <0 -> error
1462 */
1463static int write_partial_kvec(struct ceph_connection *con)
1464{
1465 int ret;
1466
1467 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1468 while (con->out_kvec_bytes > 0) {
1469 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1470 con->out_kvec_left, con->out_kvec_bytes,
1471 con->out_more);
1472 if (ret <= 0)
1473 goto out;
1474 con->out_kvec_bytes -= ret;
1475 if (con->out_kvec_bytes == 0)
1476 break; /* done */
f42299e6
AE		 1477
1478 /* account for full iov entries consumed */
1479 while (ret >= con->out_kvec_cur->iov_len) {
1480 BUG_ON(!con->out_kvec_left);
1481 ret -= con->out_kvec_cur->iov_len;
1482 con->out_kvec_cur++;
1483 con->out_kvec_left--;
1484 }
1485 /* and for a partially-consumed entry */
1486 if (ret) {
1487 con->out_kvec_cur->iov_len -= ret;
1488 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1489 }
1490 }
1491 con->out_kvec_left = 0;
1492 con->out_kvec_is_msg = false;
1493 ret = 1;
1494out:
1495 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1496 con->out_kvec_bytes, con->out_kvec_left, ret);
1497 return ret; /* done! */
1498}
1499
35b62808
AE		 1500static u32 ceph_crc32c_page(u32 crc, struct page *page,
1501 unsigned int page_offset,
1502 unsigned int length)
1503{
1504 char *kaddr;
1505
1506 kaddr = kmap(page);
1507 BUG_ON(kaddr == NULL);
1508 crc = crc32c(crc, kaddr + page_offset, length);
1509 kunmap(page);
1510
1511 return crc;
1512}
31b8006e
SW		 1513/*
1514 * Write as much message data payload as we can. If we finish, queue
1515 * up the footer.
1516 * 1 -> done, footer is now queued in out_kvec[].
1517 * 0 -> socket full, but more to do
1518 * <0 -> error
1519 */
34d2d200 1520static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1521{
1522 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1523 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1524 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1525 u32 crc;
31b8006e 1526
859a35d5 1527 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1528
5240d9f9 1529 if (list_empty(&msg->data))
4c59b4a2
AE		 1530 return -EINVAL;
1531
5821bd8c
AE		 1532 /*
1533 * Iterate through each page that contains data to be
1534 * written, and send as much as possible for each.
1535 *
1536 * If we are calculating the data crc (the default), we will
1537 * need to map the page. If we have no pages, they have
1538 * been revoked, so use the zero page.
1539 */
f5db90bc 1540 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1541 while (cursor->resid) {
8a166d05 1542 struct page *page;
e387d525
AE		 1543 size_t page_offset;
1544 size_t length;
8a166d05 1545 bool last_piece;
8ea299bc 1546 bool need_crc;
f5db90bc 1547 int ret;
68b4476b 1548
8ae4f4f5 1549 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 1550 &last_piece);
e387d525 1551 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1552 length, last_piece);
f5db90bc
AE		 1553 if (ret <= 0) {
1554 if (do_datacrc)
1555 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1556
f5db90bc
AE		 1557 return ret;
1558 }
143334ff
AE		 1559 if (do_datacrc && cursor->need_crc)
1560 crc = ceph_crc32c_page(crc, page, page_offset, length);
8ae4f4f5 1561 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
31b8006e
SW		 1562 }
1563
34d2d200 1564 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1565
1566 /* prepare and queue up footer, too */
f5db90bc
AE		 1567 if (do_datacrc)
1568 msg->footer.data_crc = cpu_to_le32(crc);
1569 else
84ca8fc8 1570 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1571 con_out_kvec_reset(con);
859eb799 1572 prepare_write_message_footer(con);
f5db90bc
AE		 1573
1574 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1575}
1576
1577/*
1578 * write some zeros
1579 */
1580static int write_partial_skip(struct ceph_connection *con)
1581{
1582 int ret;
1583
1584 while (con->out_skip > 0) {
31739139 1585 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1586
e1dcb128 1587 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1588 if (ret <= 0)
1589 goto out;
1590 con->out_skip -= ret;
1591 }
1592 ret = 1;
1593out:
1594 return ret;
1595}
1596
1597/*
1598 * Prepare to read connection handshake, or an ack.
1599 */
eed0ef2c
SW		 1600static void prepare_read_banner(struct ceph_connection *con)
1601{
1602 dout("prepare_read_banner %p\n", con);
1603 con->in_base_pos = 0;
1604}
1605
31b8006e
SW		 1606static void prepare_read_connect(struct ceph_connection *con)
1607{
1608 dout("prepare_read_connect %p\n", con);
1609 con->in_base_pos = 0;
1610}
1611
1612static void prepare_read_ack(struct ceph_connection *con)
1613{
1614 dout("prepare_read_ack %p\n", con);
1615 con->in_base_pos = 0;
1616}
1617
3a23083b
SW		 1618static void prepare_read_seq(struct ceph_connection *con)
1619{
1620 dout("prepare_read_seq %p\n", con);
1621 con->in_base_pos = 0;
1622 con->in_tag = CEPH_MSGR_TAG_SEQ;
1623}
1624
31b8006e
SW		 1625static void prepare_read_tag(struct ceph_connection *con)
1626{
1627 dout("prepare_read_tag %p\n", con);
1628 con->in_base_pos = 0;
1629 con->in_tag = CEPH_MSGR_TAG_READY;
1630}
1631
1632/*
1633 * Prepare to read a message.
1634 */
1635static int prepare_read_message(struct ceph_connection *con)
1636{
1637 dout("prepare_read_message %p\n", con);
1638 BUG_ON(con->in_msg != NULL);
1639 con->in_base_pos = 0;
1640 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1641 return 0;
1642}
1643
1644
1645static int read_partial(struct ceph_connection *con,
fd51653f 1646 int end, int size, void *object)
31b8006e 1647{
e6cee71f
AE		 1648 while (con->in_base_pos < end) {
1649 int left = end - con->in_base_pos;
31b8006e
SW		 1650 int have = size - left;
1651 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1652 if (ret <= 0)
1653 return ret;
1654 con->in_base_pos += ret;
1655 }
1656 return 1;
1657}
1658
1659
1660/*
1661 * Read all or part of the connect-side handshake on a new connection
1662 */
eed0ef2c 1663static int read_partial_banner(struct ceph_connection *con)
31b8006e 1664{
fd51653f
AE		 1665 int size;
1666 int end;
1667 int ret;
31b8006e 1668
eed0ef2c 1669 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1670
1671 /* peer's banner */
fd51653f
AE		 1672 size = strlen(CEPH_BANNER);
1673 end = size;
1674 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1675 if (ret <= 0)
1676 goto out;
fd51653f
AE		 1677
1678 size = sizeof (con->actual_peer_addr);
1679 end += size;
1680 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1681 if (ret <= 0)
1682 goto out;
fd51653f
AE		 1683
1684 size = sizeof (con->peer_addr_for_me);
1685 end += size;
1686 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1687 if (ret <= 0)
1688 goto out;
fd51653f 1689
eed0ef2c
SW		 1690out:
1691 return ret;
1692}
1693
1694static int read_partial_connect(struct ceph_connection *con)
1695{
fd51653f
AE		 1696 int size;
1697 int end;
1698 int ret;
eed0ef2c
SW		 1699
1700 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1701
fd51653f
AE		 1702 size = sizeof (con->in_reply);
1703 end = size;
1704 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1705 if (ret <= 0)
1706 goto out;
fd51653f
AE		 1707
1708 size = le32_to_cpu(con->in_reply.authorizer_len);
1709 end += size;
1710 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1711 if (ret <= 0)
1712 goto out;
31b8006e 1713
4e7a5dcd
SW		 1714 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1715 con, (int)con->in_reply.tag,
1716 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1717 le32_to_cpu(con->in_reply.global_seq));
1718out:
1719 return ret;
eed0ef2c 1720
31b8006e
SW		 1721}
1722
1723/*
1724 * Verify the hello banner looks okay.
1725 */
1726static int verify_hello(struct ceph_connection *con)
1727{
1728 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1729 pr_err("connect to %s got bad banner\n",
3d14c5d2 1730 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1731 con->error_msg = "protocol error, bad banner";
1732 return -1;
1733 }
1734 return 0;
1735}
1736
1737static bool addr_is_blank(struct sockaddr_storage *ss)
1738{
1739 switch (ss->ss_family) {
1740 case AF_INET:
1741 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1742 case AF_INET6:
1743 return
1744 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1745 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1746 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1747 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1748 }
1749 return false;
1750}
1751
1752static int addr_port(struct sockaddr_storage *ss)
1753{
1754 switch (ss->ss_family) {
1755 case AF_INET:
f28bcfbe 1756 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1757 case AF_INET6:
f28bcfbe 1758 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1759 }
1760 return 0;
1761}
1762
1763static void addr_set_port(struct sockaddr_storage *ss, int p)
1764{
1765 switch (ss->ss_family) {
1766 case AF_INET:
1767 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1768 break;
31b8006e
SW		 1769 case AF_INET6:
1770 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1771 break;
31b8006e
SW		 1772 }
1773}
1774
ee3b56f2
NW		 1775/*
1776 * Unlike other *_pton function semantics, zero indicates success.
1777 */
1778static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1779 char delim, const char **ipend)
1780{
99f0f3b2
AE		 1781 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1782 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1783
1784 memset(ss, 0, sizeof(*ss));
1785
1786 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1787 ss->ss_family = AF_INET;
1788 return 0;
1789 }
1790
1791 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1792 ss->ss_family = AF_INET6;
1793 return 0;
1794 }
1795
1796 return -EINVAL;
1797}
1798
1799/*
1800 * Extract hostname string and resolve using kernel DNS facility.
1801 */
1802#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1803static int ceph_dns_resolve_name(const char *name, size_t namelen,
1804 struct sockaddr_storage *ss, char delim, const char **ipend)
1805{
1806 const char *end, *delim_p;
1807 char *colon_p, *ip_addr = NULL;
1808 int ip_len, ret;
1809
1810 /*
1811 * The end of the hostname occurs immediately preceding the delimiter or
1812 * the port marker (':') where the delimiter takes precedence.
1813 */
1814 delim_p = memchr(name, delim, namelen);
1815 colon_p = memchr(name, ':', namelen);
1816
1817 if (delim_p && colon_p)
1818 end = delim_p < colon_p ? delim_p : colon_p;
1819 else if (!delim_p && colon_p)
1820 end = colon_p;
1821 else {
1822 end = delim_p;
1823 if (!end) /* case: hostname:/ */
1824 end = name + namelen;
1825 }
1826
1827 if (end <= name)
1828 return -EINVAL;
1829
1830 /* do dns_resolve upcall */
1831 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1832 if (ip_len > 0)
1833 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1834 else
1835 ret = -ESRCH;
1836
1837 kfree(ip_addr);
1838
1839 *ipend = end;
1840
1841 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1842 ret, ret ? "failed" : ceph_pr_addr(ss));
1843
1844 return ret;
1845}
1846#else
1847static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1848 struct sockaddr_storage *ss, char delim, const char **ipend)
1849{
1850 return -EINVAL;
1851}
1852#endif
1853
1854/*
1855 * Parse a server name (IP or hostname). If a valid IP address is not found
1856 * then try to extract a hostname to resolve using userspace DNS upcall.
1857 */
1858static int ceph_parse_server_name(const char *name, size_t namelen,
1859 struct sockaddr_storage *ss, char delim, const char **ipend)
1860{
1861 int ret;
1862
1863 ret = ceph_pton(name, namelen, ss, delim, ipend);
1864 if (ret)
1865 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1866
1867 return ret;
1868}
1869
31b8006e
SW		 1870/*
1871 * Parse an ip[:port] list into an addr array. Use the default
1872 * monitor port if a port isn't specified.
1873 */
1874int ceph_parse_ips(const char *c, const char *end,
1875 struct ceph_entity_addr *addr,
1876 int max_count, int *count)
1877{
ee3b56f2 1878 int i, ret = -EINVAL;
31b8006e
SW		 1879 const char *p = c;
1880
1881 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1882 for (i = 0; i < max_count; i++) {
1883 const char *ipend;
1884 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1885 int port;
39139f64
SW		 1886 char delim = ',';
1887
1888 if (*p == '[') {
1889 delim = ']';
1890 p++;
1891 }
31b8006e 1892
ee3b56f2
NW		 1893 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1894 if (ret)
31b8006e 1895 goto bad;
ee3b56f2
NW		 1896 ret = -EINVAL;
1897
31b8006e
SW		 1898 p = ipend;
1899
39139f64
SW		 1900 if (delim == ']') {
1901 if (*p != ']') {
1902 dout("missing matching ']'\n");
1903 goto bad;
1904 }
1905 p++;
1906 }
1907
31b8006e
SW		 1908 /* port? */
1909 if (p < end && *p == ':') {
1910 port = 0;
1911 p++;
1912 while (p < end && *p >= '0' && *p <= '9') {
1913 port = (port * 10) + (*p - '0');
1914 p++;
1915 }
f48db1e9
ID		 1916 if (port == 0)
1917 port = CEPH_MON_PORT;
1918 else if (port > 65535)
31b8006e
SW		 1919 goto bad;
1920 } else {
1921 port = CEPH_MON_PORT;
1922 }
1923
1924 addr_set_port(ss, port);
1925
3d14c5d2 1926 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1927
1928 if (p == end)
1929 break;
1930 if (*p != ',')
1931 goto bad;
1932 p++;
1933 }
1934
1935 if (p != end)
1936 goto bad;
1937
1938 if (count)
1939 *count = i + 1;
1940 return 0;
1941
1942bad:
39139f64 1943 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1944 return ret;
31b8006e 1945}
3d14c5d2 1946EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1947
eed0ef2c 1948static int process_banner(struct ceph_connection *con)
31b8006e 1949{
eed0ef2c 1950 dout("process_banner on %p\n", con);
31b8006e
SW		 1951
1952 if (verify_hello(con) < 0)
1953 return -1;
1954
63f2d211
SW		 1955 ceph_decode_addr(&con->actual_peer_addr);
1956 ceph_decode_addr(&con->peer_addr_for_me);
1957
31b8006e
SW		 1958 /*
1959 * Make sure the other end is who we wanted. note that the other
1960 * end may not yet know their ip address, so if it's 0.0.0.0, give
1961 * them the benefit of the doubt.
1962 */
103e2d3a
SW		 1963 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1964 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1965 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1966 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 1967 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1968 ceph_pr_addr(&con->peer_addr.in_addr),
1969 (int)le32_to_cpu(con->peer_addr.nonce),
1970 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1971 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1972 con->error_msg = "wrong peer at address";
31b8006e
SW		 1973 return -1;
1974 }
1975
1976 /*
1977 * did we learn our address?
1978 */
1979 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1980 int port = addr_port(&con->msgr->inst.addr.in_addr);
1981
1982 memcpy(&con->msgr->inst.addr.in_addr,
1983 &con->peer_addr_for_me.in_addr,
1984 sizeof(con->peer_addr_for_me.in_addr));
1985 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1986 encode_my_addr(con->msgr);
eed0ef2c 1987 dout("process_banner learned my addr is %s\n",
3d14c5d2 1988 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1989 }
1990
eed0ef2c
SW		 1991 return 0;
1992}
1993
1994static int process_connect(struct ceph_connection *con)
1995{
3d14c5d2
YS		 1996 u64 sup_feat = con->msgr->supported_features;
1997 u64 req_feat = con->msgr->required_features;
2b3e0c90
ID		 1998 u64 server_feat = ceph_sanitize_features(
1999 le64_to_cpu(con->in_reply.features));
0da5d703 2000 int ret;
04a419f9 2001
eed0ef2c
SW		 2002 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2003
31b8006e 2004 switch (con->in_reply.tag) {
04a419f9
SW		 2005 case CEPH_MSGR_TAG_FEATURES:
2006 pr_err("%s%lld %s feature set mismatch,"
2007 " my %llx < server's %llx, missing %llx\n",
2008 ENTITY_NAME(con->peer_name),
3d14c5d2 2009 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2010 sup_feat, server_feat, server_feat & ~sup_feat);
2011 con->error_msg = "missing required protocol features";
0fa6ebc6 2012 reset_connection(con);
04a419f9
SW		 2013 return -1;
2014
31b8006e 2015 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2016 pr_err("%s%lld %s protocol version mismatch,"
2017 " my %d != server's %d\n",
2018 ENTITY_NAME(con->peer_name),
3d14c5d2 2019 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2020 le32_to_cpu(con->out_connect.protocol_version),
2021 le32_to_cpu(con->in_reply.protocol_version));
2022 con->error_msg = "protocol version mismatch";
0fa6ebc6 2023 reset_connection(con);
31b8006e
SW		 2024 return -1;
2025
4e7a5dcd
SW		 2026 case CEPH_MSGR_TAG_BADAUTHORIZER:
2027 con->auth_retry++;
2028 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2029 con->auth_retry);
2030 if (con->auth_retry == 2) {
2031 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2032 return -1;
2033 }
6d4221b5 2034 con_out_kvec_reset(con);
e825a66d 2035 ret = prepare_write_connect(con);
0da5d703
SW		 2036 if (ret < 0)
2037 return ret;
63733a0f 2038 prepare_read_connect(con);
4e7a5dcd 2039 break;
31b8006e
SW		 2040
2041 case CEPH_MSGR_TAG_RESETSESSION:
2042 /*
2043 * If we connected with a large connect_seq but the peer
2044 * has no record of a session with us (no connection, or
2045 * connect_seq == 0), they will send RESETSESION to indicate
2046 * that they must have reset their session, and may have
2047 * dropped messages.
2048 */
2049 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2050 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2051 pr_err("%s%lld %s connection reset\n",
2052 ENTITY_NAME(con->peer_name),
3d14c5d2 2053 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2054 reset_connection(con);
6d4221b5 2055 con_out_kvec_reset(con);
5a0f8fdd
AE		 2056 ret = prepare_write_connect(con);
2057 if (ret < 0)
2058 return ret;
31b8006e
SW		 2059 prepare_read_connect(con);
2060
2061 /* Tell ceph about it. */
ec302645 2062 mutex_unlock(&con->mutex);
31b8006e
SW		 2063 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2064 if (con->ops->peer_reset)
2065 con->ops->peer_reset(con);
ec302645 2066 mutex_lock(&con->mutex);
8dacc7da 2067 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2068 return -EAGAIN;
31b8006e
SW		 2069 break;
2070
2071 case CEPH_MSGR_TAG_RETRY_SESSION:
2072 /*
2073 * If we sent a smaller connect_seq than the peer has, try
2074 * again with a larger value.
2075 */
5bdca4e0 2076 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2077 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2078 le32_to_cpu(con->in_reply.connect_seq));
2079 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2080 con_out_kvec_reset(con);
5a0f8fdd
AE		 2081 ret = prepare_write_connect(con);
2082 if (ret < 0)
2083 return ret;
31b8006e
SW		 2084 prepare_read_connect(con);
2085 break;
2086
2087 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2088 /*
2089 * If we sent a smaller global_seq than the peer has, try
2090 * again with a larger value.
2091 */
eed0ef2c 2092 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2093 con->peer_global_seq,
5bdca4e0 2094 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2095 get_global_seq(con->msgr,
5bdca4e0 2096 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2097 con_out_kvec_reset(con);
5a0f8fdd
AE		 2098 ret = prepare_write_connect(con);
2099 if (ret < 0)
2100 return ret;
31b8006e
SW		 2101 prepare_read_connect(con);
2102 break;
2103
3a23083b 2104 case CEPH_MSGR_TAG_SEQ:
31b8006e 2105 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2106 if (req_feat & ~server_feat) {
2107 pr_err("%s%lld %s protocol feature mismatch,"
2108 " my required %llx > server's %llx, need %llx\n",
2109 ENTITY_NAME(con->peer_name),
3d14c5d2 2110 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2111 req_feat, server_feat, req_feat & ~server_feat);
2112 con->error_msg = "missing required protocol features";
0fa6ebc6 2113 reset_connection(con);
04a419f9
SW		 2114 return -1;
2115 }
8dacc7da 2116
122070a2 2117 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2118 con->state = CON_STATE_OPEN;
20e55c4c 2119 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2120 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2121 con->connect_seq++;
aba558e2 2122 con->peer_features = server_feat;
31b8006e
SW		 2123 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2124 con->peer_global_seq,
2125 le32_to_cpu(con->in_reply.connect_seq),
2126 con->connect_seq);
2127 WARN_ON(con->connect_seq !=
2128 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2129
2130 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2131 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2132
85effe18 2133 con->delay = 0; /* reset backoff memory */
92ac41d0 2134
3a23083b
SW		 2135 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2136 prepare_write_seq(con);
2137 prepare_read_seq(con);
2138 } else {
2139 prepare_read_tag(con);
2140 }
31b8006e
SW		 2141 break;
2142
2143 case CEPH_MSGR_TAG_WAIT:
2144 /*
2145 * If there is a connection race (we are opening
2146 * connections to each other), one of us may just have
2147 * to WAIT. This shouldn't happen if we are the
2148 * client.
2149 */
04177882
SW		 2150 pr_err("process_connect got WAIT as client\n");
2151 con->error_msg = "protocol error, got WAIT as client";
2152 return -1;
31b8006e
SW		 2153
2154 default:
2155 pr_err("connect protocol error, will retry\n");
2156 con->error_msg = "protocol error, garbage tag during connect";
2157 return -1;
2158 }
2159 return 0;
2160}
2161
2162
2163/*
2164 * read (part of) an ack
2165 */
2166static int read_partial_ack(struct ceph_connection *con)
2167{
fd51653f
AE		 2168 int size = sizeof (con->in_temp_ack);
2169 int end = size;
31b8006e 2170
fd51653f 2171 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2172}
2173
31b8006e
SW		 2174/*
2175 * We can finally discard anything that's been acked.
2176 */
2177static void process_ack(struct ceph_connection *con)
2178{
2179 struct ceph_msg *m;
2180 u64 ack = le64_to_cpu(con->in_temp_ack);
2181 u64 seq;
2182
31b8006e
SW		 2183 while (!list_empty(&con->out_sent)) {
2184 m = list_first_entry(&con->out_sent, struct ceph_msg,
2185 list_head);
2186 seq = le64_to_cpu(m->hdr.seq);
2187 if (seq > ack)
2188 break;
2189 dout("got ack for seq %llu type %d at %p\n", seq,
2190 le16_to_cpu(m->hdr.type), m);
4cf9d544 2191 m->ack_stamp = jiffies;
31b8006e
SW		 2192 ceph_msg_remove(m);
2193 }
31b8006e
SW		 2194 prepare_read_tag(con);
2195}
2196
2197
2450418c 2198static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2199 struct kvec *section,
2200 unsigned int sec_len, u32 *crc)
2450418c 2201{
68b4476b 2202 int ret, left;
2450418c
YS		 2203
2204 BUG_ON(!section);
2205
2206 while (section->iov_len < sec_len) {
2207 BUG_ON(section->iov_base == NULL);
2208 left = sec_len - section->iov_len;
2209 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2210 section->iov_len, left);
2211 if (ret <= 0)
2212 return ret;
2213 section->iov_len += ret;
2450418c 2214 }
fe3ad593
AE		 2215 if (section->iov_len == sec_len)
2216 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2217
2450418c
YS		 2218 return 1;
2219}
31b8006e 2220
34d2d200
AE		 2221static int read_partial_msg_data(struct ceph_connection *con)
2222{
2223 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2224 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2225 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2226 struct page *page;
2227 size_t page_offset;
2228 size_t length;
f5db90bc 2229 u32 crc = 0;
34d2d200
AE		 2230 int ret;
2231
2232 BUG_ON(!msg);
5240d9f9 2233 if (list_empty(&msg->data))
4c59b4a2 2234 return -EIO;
34d2d200 2235
f5db90bc
AE		 2236 if (do_datacrc)
2237 crc = con->in_data_crc;
643c68a4 2238 while (cursor->resid) {
8ae4f4f5 2239 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 2240 NULL);
686be208 2241 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2242 if (ret <= 0) {
2243 if (do_datacrc)
2244 con->in_data_crc = crc;
2245
686be208 2246 return ret;
f5db90bc 2247 }
686be208
AE		 2248
2249 if (do_datacrc)
f5db90bc 2250 crc = ceph_crc32c_page(crc, page, page_offset, ret);
8ae4f4f5 2251 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
34d2d200 2252 }
f5db90bc
AE		 2253 if (do_datacrc)
2254 con->in_data_crc = crc;
34d2d200
AE		 2255
2256 return 1; /* must return > 0 to indicate success */
2257}
2258
31b8006e
SW		 2259/*
2260 * read (part of) a message.
2261 */
686be208
AE		 2262static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2263
31b8006e
SW		 2264static int read_partial_message(struct ceph_connection *con)
2265{
2266 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2267 int size;
2268 int end;
31b8006e 2269 int ret;
95c96174 2270 unsigned int front_len, middle_len, data_len;
37675b0f 2271 bool do_datacrc = !con->msgr->nocrc;
33d07337 2272 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2273 u64 seq;
fe3ad593 2274 u32 crc;
31b8006e
SW		 2275
2276 dout("read_partial_message con %p msg %p\n", con, m);
2277
2278 /* header */
fd51653f
AE		 2279 size = sizeof (con->in_hdr);
2280 end = size;
2281 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2282 if (ret <= 0)
2283 return ret;
fe3ad593
AE		 2284
2285 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2286 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2287 pr_err("read_partial_message bad hdr "
2288 " crc %u != expected %u\n",
2289 crc, con->in_hdr.crc);
2290 return -EBADMSG;
2291 }
2292
31b8006e
SW		 2293 front_len = le32_to_cpu(con->in_hdr.front_len);
2294 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2295 return -EIO;
2296 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2297 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2298 return -EIO;
2299 data_len = le32_to_cpu(con->in_hdr.data_len);
2300 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2301 return -EIO;
2302
ae18756b
SW		 2303 /* verify seq# */
2304 seq = le64_to_cpu(con->in_hdr.seq);
2305 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2306 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2307 ENTITY_NAME(con->peer_name),
3d14c5d2 2308 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2309 seq, con->in_seq + 1);
2310 con->in_base_pos = -front_len - middle_len - data_len -
2311 sizeof(m->footer);
2312 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2313 return 0;
2314 } else if ((s64)seq - (s64)con->in_seq > 1) {
2315 pr_err("read_partial_message bad seq %lld expected %lld\n",
2316 seq, con->in_seq + 1);
2317 con->error_msg = "bad message sequence # for incoming message";
2318 return -EBADMSG;
2319 }
2320
31b8006e
SW		 2321 /* allocate message? */
2322 if (!con->in_msg) {
4740a623
SW		 2323 int skip = 0;
2324
31b8006e 2325 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2326 front_len, data_len);
4740a623
SW		 2327 ret = ceph_con_in_msg_alloc(con, &skip);
2328 if (ret < 0)
2329 return ret;
f759ebb9
AE		 2330
2331 BUG_ON(!con->in_msg ^ skip);
2332 if (con->in_msg && data_len > con->in_msg->data_length) {
b9a67899 2333 pr_warn("%s skipping long message (%u > %zd)\n",
f759ebb9
AE		 2334 __func__, data_len, con->in_msg->data_length);
2335 ceph_msg_put(con->in_msg);
2336 con->in_msg = NULL;
2337 skip = 1;
2338 }
2450418c 2339 if (skip) {
31b8006e 2340 /* skip this message */
a79832f2 2341 dout("alloc_msg said skip message\n");
31b8006e
SW		 2342 con->in_base_pos = -front_len - middle_len - data_len -
2343 sizeof(m->footer);
2344 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2345 con->in_seq++;
31b8006e
SW		 2346 return 0;
2347 }
38941f80 2348
4740a623 2349 BUG_ON(!con->in_msg);
38941f80 2350 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2351 m = con->in_msg;
2352 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2353 if (m->middle)
2354 m->middle->vec.iov_len = 0;
9d7f0f13 2355
78625051 2356 /* prepare for data payload, if any */
a4107026 2357
78625051 2358 if (data_len)
98fa5dd8 2359 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2360 }
2361
2362 /* front */
2450418c
YS		 2363 ret = read_partial_message_section(con, &m->front, front_len,
2364 &con->in_front_crc);
2365 if (ret <= 0)
2366 return ret;
31b8006e
SW		 2367
2368 /* middle */
2450418c 2369 if (m->middle) {
213c99ee
SW		 2370 ret = read_partial_message_section(con, &m->middle->vec,
2371 middle_len,
2450418c 2372 &con->in_middle_crc);
31b8006e
SW		 2373 if (ret <= 0)
2374 return ret;
31b8006e
SW		 2375 }
2376
2377 /* (page) data */
34d2d200
AE		 2378 if (data_len) {
2379 ret = read_partial_msg_data(con);
2380 if (ret <= 0)
2381 return ret;
31b8006e
SW		 2382 }
2383
31b8006e 2384 /* footer */
33d07337
YZ		 2385 if (need_sign)
2386 size = sizeof(m->footer);
2387 else
2388 size = sizeof(m->old_footer);
2389
fd51653f
AE		 2390 end += size;
2391 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2392 if (ret <= 0)
2393 return ret;
2394
33d07337
YZ		 2395 if (!need_sign) {
2396 m->footer.flags = m->old_footer.flags;
2397 m->footer.sig = 0;
2398 }
2399
31b8006e
SW		 2400 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2401 m, front_len, m->footer.front_crc, middle_len,
2402 m->footer.middle_crc, data_len, m->footer.data_crc);
2403
2404 /* crc ok? */
2405 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2406 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2407 m, con->in_front_crc, m->footer.front_crc);
2408 return -EBADMSG;
2409 }
2410 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2411 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2412 m, con->in_middle_crc, m->footer.middle_crc);
2413 return -EBADMSG;
2414 }
bca064d2 2415 if (do_datacrc &&
31b8006e
SW		 2416 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2417 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2418 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2419 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2420 return -EBADMSG;
2421 }
2422
33d07337
YZ		 2423 if (need_sign && con->ops->check_message_signature &&
2424 con->ops->check_message_signature(con, m)) {
2425 pr_err("read_partial_message %p signature check failed\n", m);
2426 return -EBADMSG;
2427 }
2428
31b8006e
SW		 2429 return 1; /* done! */
2430}
2431
2432/*
2433 * Process message. This happens in the worker thread. The callback should
2434 * be careful not to do anything that waits on other incoming messages or it
2435 * may deadlock.
2436 */
2437static void process_message(struct ceph_connection *con)
2438{
5e095e8b 2439 struct ceph_msg *msg;
31b8006e 2440
38941f80
AE		 2441 BUG_ON(con->in_msg->con != con);
2442 con->in_msg->con = NULL;
5e095e8b 2443 msg = con->in_msg;
31b8006e 2444 con->in_msg = NULL;
36eb71aa 2445 con->ops->put(con);
31b8006e
SW		 2446
2447 /* if first message, set peer_name */
2448 if (con->peer_name.type == 0)
dbad185d 2449 con->peer_name = msg->hdr.src;
31b8006e 2450
31b8006e 2451 con->in_seq++;
ec302645 2452 mutex_unlock(&con->mutex);
31b8006e
SW		 2453
2454 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2455 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2456 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2457 le16_to_cpu(msg->hdr.type),
2458 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2459 le32_to_cpu(msg->hdr.front_len),
2460 le32_to_cpu(msg->hdr.data_len),
2461 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2462 con->ops->dispatch(con, msg);
ec302645
SW		 2463
2464 mutex_lock(&con->mutex);
31b8006e
SW		 2465}
2466
2467
2468/*
2469 * Write something to the socket. Called in a worker thread when the
2470 * socket appears to be writeable and we have something ready to send.
2471 */
2472static int try_write(struct ceph_connection *con)
2473{
31b8006e
SW		 2474 int ret = 1;
2475
d59315ca 2476 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2477
31b8006e
SW		 2478more:
2479 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2480
2481 /* open the socket first? */
8dacc7da
SW		 2482 if (con->state == CON_STATE_PREOPEN) {
2483 BUG_ON(con->sock);
2484 con->state = CON_STATE_CONNECTING;
a5988c49 2485
e2200423 2486 con_out_kvec_reset(con);
e825a66d 2487 prepare_write_banner(con);
eed0ef2c 2488 prepare_read_banner(con);
31b8006e 2489
cf3e5c40 2490 BUG_ON(con->in_msg);
31b8006e
SW		 2491 con->in_tag = CEPH_MSGR_TAG_READY;
2492 dout("try_write initiating connect on %p new state %lu\n",
2493 con, con->state);
41617d0c
AE		 2494 ret = ceph_tcp_connect(con);
2495 if (ret < 0) {
31b8006e 2496 con->error_msg = "connect error";
31b8006e
SW		 2497 goto out;
2498 }
2499 }
2500
2501more_kvec:
2502 /* kvec data queued? */
2503 if (con->out_skip) {
2504 ret = write_partial_skip(con);
2505 if (ret <= 0)
42961d23 2506 goto out;
31b8006e
SW		 2507 }
2508 if (con->out_kvec_left) {
2509 ret = write_partial_kvec(con);
2510 if (ret <= 0)
42961d23 2511 goto out;
31b8006e
SW		 2512 }
2513
2514 /* msg pages? */
2515 if (con->out_msg) {
c86a2930
SW		 2516 if (con->out_msg_done) {
2517 ceph_msg_put(con->out_msg);
2518 con->out_msg = NULL; /* we're done with this one */
2519 goto do_next;
2520 }
2521
34d2d200 2522 ret = write_partial_message_data(con);
31b8006e
SW		 2523 if (ret == 1)
2524 goto more_kvec; /* we need to send the footer, too! */
2525 if (ret == 0)
42961d23 2526 goto out;
31b8006e 2527 if (ret < 0) {
34d2d200 2528 dout("try_write write_partial_message_data err %d\n",
31b8006e 2529 ret);
42961d23 2530 goto out;
31b8006e
SW		 2531 }
2532 }
2533
c86a2930 2534do_next:
8dacc7da 2535 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2536 /* is anything else pending? */
2537 if (!list_empty(&con->out_queue)) {
2538 prepare_write_message(con);
2539 goto more;
2540 }
2541 if (con->in_seq > con->in_seq_acked) {
2542 prepare_write_ack(con);
2543 goto more;
2544 }
c9ffc77a 2545 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2546 prepare_write_keepalive(con);
2547 goto more;
2548 }
2549 }
2550
2551 /* Nothing to do! */
c9ffc77a 2552 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2553 dout("try_write nothing else to write.\n");
31b8006e
SW		 2554 ret = 0;
2555out:
42961d23 2556 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2557 return ret;
2558}
2559
2560
2561
2562/*
2563 * Read what we can from the socket.
2564 */
2565static int try_read(struct ceph_connection *con)
2566{
31b8006e
SW		 2567 int ret = -1;
2568
8dacc7da
SW		 2569more:
2570 dout("try_read start on %p state %lu\n", con, con->state);
2571 if (con->state != CON_STATE_CONNECTING &&
2572 con->state != CON_STATE_NEGOTIATING &&
2573 con->state != CON_STATE_OPEN)
31b8006e
SW		 2574 return 0;
2575
8dacc7da 2576 BUG_ON(!con->sock);
ec302645 2577
31b8006e
SW		 2578 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2579 con->in_base_pos);
0da5d703 2580
8dacc7da 2581 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2582 dout("try_read connecting\n");
2583 ret = read_partial_banner(con);
2584 if (ret <= 0)
ab166d5a 2585 goto out;
7593af92
AE		 2586 ret = process_banner(con);
2587 if (ret < 0)
2588 goto out;
2589
8dacc7da 2590 con->state = CON_STATE_NEGOTIATING;
7593af92 2591
6d4221b5
JS		 2592 /*
2593 * Received banner is good, exchange connection info.
2594 * Do not reset out_kvec, as sending our banner raced
2595 * with receiving peer banner after connect completed.
2596 */
7593af92
AE		 2597 ret = prepare_write_connect(con);
2598 if (ret < 0)
2599 goto out;
2600 prepare_read_connect(con);
2601
2602 /* Send connection info before awaiting response */
0da5d703
SW		 2603 goto out;
2604 }
2605
8dacc7da 2606 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2607 dout("try_read negotiating\n");
31b8006e
SW		 2608 ret = read_partial_connect(con);
2609 if (ret <= 0)
31b8006e 2610 goto out;
98bdb0aa
SW		 2611 ret = process_connect(con);
2612 if (ret < 0)
2613 goto out;
31b8006e
SW		 2614 goto more;
2615 }
2616
122070a2 2617 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2618
31b8006e
SW		 2619 if (con->in_base_pos < 0) {
2620 /*
2621 * skipping + discarding content.
2622 *
2623 * FIXME: there must be a better way to do this!
2624 */
84495f49
AE		 2625 static char buf[SKIP_BUF_SIZE];
2626 int skip = min((int) sizeof (buf), -con->in_base_pos);
2627
31b8006e
SW		 2628 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2629 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2630 if (ret <= 0)
98bdb0aa 2631 goto out;
31b8006e
SW		 2632 con->in_base_pos += ret;
2633 if (con->in_base_pos)
2634 goto more;
2635 }
2636 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2637 /*
2638 * what's next?
2639 */
2640 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2641 if (ret <= 0)
98bdb0aa 2642 goto out;
31b8006e
SW		 2643 dout("try_read got tag %d\n", (int)con->in_tag);
2644 switch (con->in_tag) {
2645 case CEPH_MSGR_TAG_MSG:
2646 prepare_read_message(con);
2647 break;
2648 case CEPH_MSGR_TAG_ACK:
2649 prepare_read_ack(con);
2650 break;
2651 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2652 con_close_socket(con);
2653 con->state = CON_STATE_CLOSED;
98bdb0aa 2654 goto out;
31b8006e
SW		 2655 default:
2656 goto bad_tag;
2657 }
2658 }
2659 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2660 ret = read_partial_message(con);
2661 if (ret <= 0) {
2662 switch (ret) {
2663 case -EBADMSG:
2664 con->error_msg = "bad crc";
2665 ret = -EIO;
98bdb0aa 2666 break;
31b8006e
SW		 2667 case -EIO:
2668 con->error_msg = "io error";
98bdb0aa 2669 break;
31b8006e 2670 }
98bdb0aa 2671 goto out;
31b8006e
SW		 2672 }
2673 if (con->in_tag == CEPH_MSGR_TAG_READY)
2674 goto more;
2675 process_message(con);
7b862e07
SW		 2676 if (con->state == CON_STATE_OPEN)
2677 prepare_read_tag(con);
31b8006e
SW		 2678 goto more;
2679 }
3a23083b
SW		 2680 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2681 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2682 /*
2683 * the final handshake seq exchange is semantically
2684 * equivalent to an ACK
2685 */
31b8006e
SW		 2686 ret = read_partial_ack(con);
2687 if (ret <= 0)
98bdb0aa 2688 goto out;
31b8006e
SW		 2689 process_ack(con);
2690 goto more;
2691 }
2692
31b8006e 2693out:
98bdb0aa 2694 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2695 return ret;
2696
2697bad_tag:
2698 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2699 con->error_msg = "protocol error, garbage tag";
2700 ret = -1;
2701 goto out;
2702}
2703
2704
2705/*
802c6d96
AE		 2706 * Atomically queue work on a connection after the specified delay.
2707 * Bump @con reference to avoid races with connection teardown.
2708 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2709 */
802c6d96 2710static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2711{
31b8006e 2712 if (!con->ops->get(con)) {
802c6d96 2713 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2714 return -ENOENT;
31b8006e
SW		 2715 }
2716
802c6d96
AE		 2717 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2718 dout("%s %p - already queued\n", __func__, con);
31b8006e 2719 con->ops->put(con);
802c6d96 2720 return -EBUSY;
31b8006e 2721 }
802c6d96
AE		 2722
2723 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2724 return 0;
2725}
2726
2727static void queue_con(struct ceph_connection *con)
2728{
2729 (void) queue_con_delay(con, 0);
31b8006e
SW		 2730}
2731
37ab77ac
ID		 2732static void cancel_con(struct ceph_connection *con)
2733{
2734 if (cancel_delayed_work(&con->work)) {
2735 dout("%s %p\n", __func__, con);
2736 con->ops->put(con);
2737 }
2738}
2739
7bb21d68
AE		 2740static bool con_sock_closed(struct ceph_connection *con)
2741{
c9ffc77a 2742 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2743 return false;
2744
2745#define CASE(x) \
2746 case CON_STATE_ ## x: \
2747 con->error_msg = "socket closed (con state " #x ")"; \
2748 break;
2749
2750 switch (con->state) {
2751 CASE(CLOSED);
2752 CASE(PREOPEN);
2753 CASE(CONNECTING);
2754 CASE(NEGOTIATING);
2755 CASE(OPEN);
2756 CASE(STANDBY);
2757 default:
b9a67899 2758 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2759 __func__, con, con->state);
2760 con->error_msg = "unrecognized con state";
2761 BUG();
2762 break;
2763 }
2764#undef CASE
2765
2766 return true;
2767}
2768
f20a39fd
AE		 2769static bool con_backoff(struct ceph_connection *con)
2770{
2771 int ret;
2772
2773 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2774 return false;
2775
2776 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2777 if (ret) {
2778 dout("%s: con %p FAILED to back off %lu\n", __func__,
2779 con, con->delay);
2780 BUG_ON(ret == -ENOENT);
2781 con_flag_set(con, CON_FLAG_BACKOFF);
2782 }
2783
2784 return true;
2785}
2786
93209264
AE		 2787/* Finish fault handling; con->mutex must *not* be held here */
2788
2789static void con_fault_finish(struct ceph_connection *con)
2790{
2791 /*
2792 * in case we faulted due to authentication, invalidate our
2793 * current tickets so that we can get new ones.
2794 */
2795 if (con->auth_retry && con->ops->invalidate_authorizer) {
2796 dout("calling invalidate_authorizer()\n");
2797 con->ops->invalidate_authorizer(con);
2798 }
2799
2800 if (con->ops->fault)
2801 con->ops->fault(con);
2802}
2803
31b8006e
SW		 2804/*
2805 * Do some work on a connection. Drop a connection ref when we're done.
2806 */
2807static void con_work(struct work_struct *work)
2808{
2809 struct ceph_connection *con = container_of(work, struct ceph_connection,
2810 work.work);
89baaa57 2811 unsigned long pflags = current->flags;
49659416 2812 bool fault;
31b8006e 2813
89baaa57
MC		 2814 current->flags |= PF_MEMALLOC;
2815
9dd4658d 2816 mutex_lock(&con->mutex);
49659416
AE		 2817 while (true) {
2818 int ret;
31b8006e 2819
49659416
AE		 2820 if ((fault = con_sock_closed(con))) {
2821 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2822 break;
2823 }
2824 if (con_backoff(con)) {
2825 dout("%s: con %p BACKOFF\n", __func__, con);
2826 break;
2827 }
2828 if (con->state == CON_STATE_STANDBY) {
2829 dout("%s: con %p STANDBY\n", __func__, con);
2830 break;
2831 }
2832 if (con->state == CON_STATE_CLOSED) {
2833 dout("%s: con %p CLOSED\n", __func__, con);
2834 BUG_ON(con->sock);
2835 break;
2836 }
2837 if (con->state == CON_STATE_PREOPEN) {
2838 dout("%s: con %p PREOPEN\n", __func__, con);
2839 BUG_ON(con->sock);
2840 }
0da5d703 2841
49659416
AE		 2842 ret = try_read(con);
2843 if (ret < 0) {
2844 if (ret == -EAGAIN)
2845 continue;
2846 con->error_msg = "socket error on read";
2847 fault = true;
2848 break;
2849 }
2850
2851 ret = try_write(con);
2852 if (ret < 0) {
2853 if (ret == -EAGAIN)
2854 continue;
2855 con->error_msg = "socket error on write";
2856 fault = true;
2857 }
2858
2859 break; /* If we make it to here, we're done */
3a140a0d 2860 }
b6e7b6a1
AE		 2861 if (fault)
2862 con_fault(con);
9dd4658d 2863 mutex_unlock(&con->mutex);
0da5d703 2864
b6e7b6a1
AE		 2865 if (fault)
2866 con_fault_finish(con);
2867
2868 con->ops->put(con);
89baaa57
MC		 2869
2870 tsk_restore_flags(current, pflags, PF_MEMALLOC);
31b8006e
SW		 2871}
2872
31b8006e
SW		 2873/*
2874 * Generic error/fault handler. A retry mechanism is used with
2875 * exponential backoff
2876 */
93209264 2877static void con_fault(struct ceph_connection *con)
31b8006e 2878{
b9a67899
JP		 2879 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2880 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2881 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2882 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2883
122070a2 2884 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2885 con->state != CON_STATE_NEGOTIATING &&
2886 con->state != CON_STATE_OPEN);
ec302645 2887
31b8006e 2888 con_close_socket(con);
5e095e8b 2889
c9ffc77a 2890 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2891 dout("fault on LOSSYTX channel, marking CLOSED\n");
2892 con->state = CON_STATE_CLOSED;
93209264 2893 return;
3b5ede07
SW		 2894 }
2895
5e095e8b 2896 if (con->in_msg) {
38941f80
AE		 2897 BUG_ON(con->in_msg->con != con);
2898 con->in_msg->con = NULL;
5e095e8b
SW		 2899 ceph_msg_put(con->in_msg);
2900 con->in_msg = NULL;
36eb71aa 2901 con->ops->put(con);
5e095e8b 2902 }
31b8006e 2903
e80a52d1
SW		 2904 /* Requeue anything that hasn't been acked */
2905 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2906
e76661d0
SW		 2907 /* If there are no messages queued or keepalive pending, place
2908 * the connection in a STANDBY state */
2909 if (list_empty(&con->out_queue) &&
c9ffc77a 2910 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2911 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2912 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2913 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2914 } else {
2915 /* retry after a delay. */
8dacc7da 2916 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2917 if (con->delay == 0)
2918 con->delay = BASE_DELAY_INTERVAL;
2919 else if (con->delay < MAX_DELAY_INTERVAL)
2920 con->delay *= 2;
c9ffc77a 2921 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2922 queue_con(con);
31b8006e 2923 }
31b8006e
SW		 2924}
2925
2926
2927
2928/*
15d9882c 2929 * initialize a new messenger instance
31b8006e 2930 */
15d9882c
AE		 2931void ceph_messenger_init(struct ceph_messenger *msgr,
2932 struct ceph_entity_addr *myaddr,
12b4629a
ID		 2933 u64 supported_features,
2934 u64 required_features,
ba988f87
CH		 2935 bool nocrc,
2936 bool tcp_nodelay)
31b8006e 2937{
3d14c5d2
YS		 2938 msgr->supported_features = supported_features;
2939 msgr->required_features = required_features;
2940
31b8006e
SW		 2941 spin_lock_init(&msgr->global_seq_lock);
2942
31b8006e
SW		 2943 if (myaddr)
2944 msgr->inst.addr = *myaddr;
2945
2946 /* select a random nonce */
ac8839d7 2947 msgr->inst.addr.type = 0;
103e2d3a 2948 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2949 encode_my_addr(msgr);
15d9882c 2950 msgr->nocrc = nocrc;
ba988f87 2951 msgr->tcp_nodelay = tcp_nodelay;
31b8006e 2952
a2a32584 2953 atomic_set(&msgr->stopping, 0);
31b8006e 2954
15d9882c 2955 dout("%s %p\n", __func__, msgr);
31b8006e 2956}
15d9882c 2957EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2958
e00de341
SW		 2959static void clear_standby(struct ceph_connection *con)
2960{
2961 /* come back from STANDBY? */
8dacc7da 2962 if (con->state == CON_STATE_STANDBY) {
e00de341 2963 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2964 con->state = CON_STATE_PREOPEN;
e00de341 2965 con->connect_seq++;
c9ffc77a
AE		 2966 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2967 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2968 }
2969}
2970
31b8006e
SW		 2971/*
2972 * Queue up an outgoing message on the given connection.
2973 */
2974void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2975{
31b8006e 2976 /* set src+dst */
dbad185d 2977 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2978 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2979 msg->needs_out_seq = true;
2980
ec302645 2981 mutex_lock(&con->mutex);
92ce034b 2982
8dacc7da 2983 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2984 dout("con_send %p closed, dropping %p\n", con, msg);
2985 ceph_msg_put(msg);
2986 mutex_unlock(&con->mutex);
2987 return;
2988 }
2989
38941f80 2990 BUG_ON(msg->con != NULL);
36eb71aa 2991 msg->con = con->ops->get(con);
92ce034b
AE		 2992 BUG_ON(msg->con == NULL);
2993
31b8006e
SW		 2994 BUG_ON(!list_empty(&msg->list_head));
2995 list_add_tail(&msg->list_head, &con->out_queue);
2996 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2997 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2998 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2999 le32_to_cpu(msg->hdr.front_len),
3000 le32_to_cpu(msg->hdr.middle_len),
3001 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3002
3003 clear_standby(con);
ec302645 3004 mutex_unlock(&con->mutex);
31b8006e
SW		 3005
3006 /* if there wasn't anything waiting to send before, queue
3007 * new work */
c9ffc77a 3008 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3009 queue_con(con);
3010}
3d14c5d2 3011EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3012
3013/*
3014 * Revoke a message that was previously queued for send
3015 */
6740a845 3016void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3017{
6740a845
AE		 3018 struct ceph_connection *con = msg->con;
3019
3020 if (!con)
3021 return; /* Message not in our possession */
3022
ec302645 3023 mutex_lock(&con->mutex);
31b8006e 3024 if (!list_empty(&msg->list_head)) {
38941f80 3025 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3026 list_del_init(&msg->list_head);
38941f80 3027 BUG_ON(msg->con == NULL);
36eb71aa 3028 msg->con->ops->put(msg->con);
38941f80 3029 msg->con = NULL;
31b8006e 3030 msg->hdr.seq = 0;
38941f80 3031
31b8006e 3032 ceph_msg_put(msg);
ed98adad
SW		 3033 }
3034 if (con->out_msg == msg) {
38941f80 3035 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 3036 con->out_msg = NULL;
31b8006e
SW		 3037 if (con->out_kvec_is_msg) {
3038 con->out_skip = con->out_kvec_bytes;
3039 con->out_kvec_is_msg = false;
3040 }
ed98adad 3041 msg->hdr.seq = 0;
92ce034b
AE		 3042
3043 ceph_msg_put(msg);
31b8006e 3044 }
ec302645 3045 mutex_unlock(&con->mutex);
31b8006e
SW		 3046}
3047
350b1c32 3048/*
0d59ab81 3049 * Revoke a message that we may be reading data into
350b1c32 3050 */
8921d114 3051void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3052{
8921d114
AE		 3053 struct ceph_connection *con;
3054
3055 BUG_ON(msg == NULL);
3056 if (!msg->con) {
3057 dout("%s msg %p null con\n", __func__, msg);
3058
3059 return; /* Message not in our possession */
3060 }
3061
3062 con = msg->con;
350b1c32 3063 mutex_lock(&con->mutex);
8921d114 3064 if (con->in_msg == msg) {
95c96174
ED		 3065 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3066 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3067 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3068
3069 /* skip rest of message */
8921d114
AE		 3070 dout("%s %p msg %p revoked\n", __func__, con, msg);
3071 con->in_base_pos = con->in_base_pos -
350b1c32 3072 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3073 front_len -
3074 middle_len -
3075 data_len -
350b1c32 3076 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3077 ceph_msg_put(con->in_msg);
3078 con->in_msg = NULL;
3079 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3080 con->in_seq++;
350b1c32 3081 } else {
8921d114
AE		 3082 dout("%s %p in_msg %p msg %p no-op\n",
3083 __func__, con, con->in_msg, msg);
350b1c32
SW		 3084 }
3085 mutex_unlock(&con->mutex);
3086}
3087
31b8006e
SW		 3088/*
3089 * Queue a keepalive byte to ensure the tcp connection is alive.
3090 */
3091void ceph_con_keepalive(struct ceph_connection *con)
3092{
e00de341 3093 dout("con_keepalive %p\n", con);
00650931 3094 mutex_lock(&con->mutex);
e00de341 3095 clear_standby(con);
00650931 3096 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3097 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3098 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3099 queue_con(con);
3100}
3d14c5d2 3101EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3102
6644ed7b 3103static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3104{
6644ed7b
AE		 3105 struct ceph_msg_data *data;
3106
3107 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3108 return NULL;
3109
81b36be4 3110 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3111 if (data)
3112 data->type = type;
5240d9f9 3113 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3114
3115 return data;
3116}
3117
3118static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3119{
3120 if (!data)
3121 return;
3122
5240d9f9 3123 WARN_ON(!list_empty(&data->links));
e4339d28 3124 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3125 ceph_pagelist_release(data->pagelist);
81b36be4 3126 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3127}
3128
90af3602 3129void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3130 size_t length, size_t alignment)
02afca6c 3131{
6644ed7b
AE		 3132 struct ceph_msg_data *data;
3133
07aa1558
AE		 3134 BUG_ON(!pages);
3135 BUG_ON(!length);
6644ed7b
AE		 3136
3137 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3138 BUG_ON(!data);
3139 data->pages = pages;
3140 data->length = length;
3141 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3142
5240d9f9
AE		 3143 list_add_tail(&data->links, &msg->data);
3144 msg->data_length += length;
02afca6c 3145}
90af3602 3146EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3147
90af3602 3148void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3149 struct ceph_pagelist *pagelist)
3150{
6644ed7b
AE		 3151 struct ceph_msg_data *data;
3152
07aa1558
AE		 3153 BUG_ON(!pagelist);
3154 BUG_ON(!pagelist->length);
27fa8385 3155
6644ed7b
AE		 3156 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3157 BUG_ON(!data);
3158 data->pagelist = pagelist;
3159
5240d9f9
AE		 3160 list_add_tail(&data->links, &msg->data);
3161 msg->data_length += pagelist->length;
27fa8385 3162}
90af3602 3163EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3164
ea96571f 3165#ifdef CONFIG_BLOCK
90af3602 3166void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3167 size_t length)
27fa8385 3168{
6644ed7b
AE		 3169 struct ceph_msg_data *data;
3170
07aa1558 3171 BUG_ON(!bio);
27fa8385 3172
6644ed7b
AE		 3173 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3174 BUG_ON(!data);
3175 data->bio = bio;
c851c495 3176 data->bio_length = length;
6644ed7b 3177
5240d9f9
AE		 3178 list_add_tail(&data->links, &msg->data);
3179 msg->data_length += length;
27fa8385 3180}
90af3602 3181EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3182#endif /* CONFIG_BLOCK */
27fa8385 3183
31b8006e
SW		 3184/*
3185 * construct a new message with given type, size
3186 * the new msg has a ref count of 1.
3187 */
b61c2763
SW		 3188struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3189 bool can_fail)
31b8006e
SW		 3190{
3191 struct ceph_msg *m;
3192
e3d5d638 3193 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3194 if (m == NULL)
3195 goto out;
31b8006e
SW		 3196
3197 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3198 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3199 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3200
9516e45b
AE		 3201 INIT_LIST_HEAD(&m->list_head);
3202 kref_init(&m->kref);
5240d9f9 3203 INIT_LIST_HEAD(&m->data);
ca20892d 3204
31b8006e
SW		 3205 /* front */
3206 if (front_len) {
eeb0bed5 3207 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3208 if (m->front.iov_base == NULL) {
b61c2763 3209 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3210 front_len);
3211 goto out2;
3212 }
3213 } else {
3214 m->front.iov_base = NULL;
3215 }
f2be82b0 3216 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3217
bb257664 3218 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3219 return m;
3220
3221out2:
3222 ceph_msg_put(m);
3223out:
b61c2763
SW		 3224 if (!can_fail) {
3225 pr_err("msg_new can't create type %d front %d\n", type,
3226 front_len);
f0ed1b7c 3227 WARN_ON(1);
b61c2763
SW		 3228 } else {
3229 dout("msg_new can't create type %d front %d\n", type,
3230 front_len);
3231 }
a79832f2 3232 return NULL;
31b8006e 3233}
3d14c5d2 3234EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3235
31b8006e
SW		 3236/*
3237 * Allocate "middle" portion of a message, if it is needed and wasn't
3238 * allocated by alloc_msg. This allows us to read a small fixed-size
3239 * per-type header in the front and then gracefully fail (i.e.,
3240 * propagate the error to the caller based on info in the front) when
3241 * the middle is too large.
3242 */
2450418c 3243static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3244{
3245 int type = le16_to_cpu(msg->hdr.type);
3246 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3247
3248 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3249 ceph_msg_type_name(type), middle_len);
3250 BUG_ON(!middle_len);
3251 BUG_ON(msg->middle);
3252
b6c1d5b8 3253 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3254 if (!msg->middle)
3255 return -ENOMEM;
3256 return 0;
3257}
3258
2450418c 3259/*
1c20f2d2
AE		 3260 * Allocate a message for receiving an incoming message on a
3261 * connection, and save the result in con->in_msg. Uses the
3262 * connection's private alloc_msg op if available.
3263 *
4740a623
SW		 3264 * Returns 0 on success, or a negative error code.
3265 *
3266 * On success, if we set *skip = 1:
3267 * - the next message should be skipped and ignored.
3268 * - con->in_msg == NULL
3269 * or if we set *skip = 0:
3270 * - con->in_msg is non-null.
3271 * On error (ENOMEM, EAGAIN, ...),
3272 * - con->in_msg == NULL
2450418c 3273 */
4740a623 3274static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3275{
4740a623 3276 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3277 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3278 struct ceph_msg *msg;
4740a623 3279 int ret = 0;
2450418c 3280
1c20f2d2 3281 BUG_ON(con->in_msg != NULL);
53ded495 3282 BUG_ON(!con->ops->alloc_msg);
2450418c 3283
53ded495
AE		 3284 mutex_unlock(&con->mutex);
3285 msg = con->ops->alloc_msg(con, hdr, skip);
3286 mutex_lock(&con->mutex);
3287 if (con->state != CON_STATE_OPEN) {
3288 if (msg)
1d866d1c 3289 ceph_msg_put(msg);
53ded495
AE		 3290 return -EAGAIN;
3291 }
4137577a
AE		 3292 if (msg) {
3293 BUG_ON(*skip);
3294 con->in_msg = msg;
36eb71aa 3295 con->in_msg->con = con->ops->get(con);
92ce034b 3296 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3297 } else {
3298 /*
3299 * Null message pointer means either we should skip
3300 * this message or we couldn't allocate memory. The
3301 * former is not an error.
3302 */
3303 if (*skip)
3304 return 0;
3305 con->error_msg = "error allocating memory for incoming message";
3306
53ded495 3307 return -ENOMEM;
2450418c 3308 }
1c20f2d2 3309 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3310
1c20f2d2
AE		 3311 if (middle_len && !con->in_msg->middle) {
3312 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3313 if (ret < 0) {
1c20f2d2
AE		 3314 ceph_msg_put(con->in_msg);
3315 con->in_msg = NULL;
2450418c
YS		 3316 }
3317 }
9d7f0f13 3318
4740a623 3319 return ret;
2450418c
YS		 3320}
3321
31b8006e
SW		 3322
3323/*
3324 * Free a generically kmalloc'd message.
3325 */
0215e44b 3326static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3327{
0215e44b 3328 dout("%s %p\n", __func__, m);
4965fc38 3329 kvfree(m->front.iov_base);
e3d5d638 3330 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3331}
3332
0215e44b 3333static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3334{
3335 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3336 LIST_HEAD(data);
3337 struct list_head *links;
3338 struct list_head *next;
31b8006e 3339
0215e44b 3340 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3341 WARN_ON(!list_empty(&m->list_head));
3342
3343 /* drop middle, data, if any */
3344 if (m->middle) {
3345 ceph_buffer_put(m->middle);
3346 m->middle = NULL;
31b8006e 3347 }
5240d9f9
AE		 3348
3349 list_splice_init(&m->data, &data);
3350 list_for_each_safe(links, next, &data) {
3351 struct ceph_msg_data *data;
3352
3353 data = list_entry(links, struct ceph_msg_data, links);
3354 list_del_init(links);
3355 ceph_msg_data_destroy(data);
3356 }
a1930804 3357 m->data_length = 0;
58bb3b37 3358
c2e552e7
SW		 3359 if (m->pool)
3360 ceph_msgpool_put(m->pool, m);
3361 else
0215e44b
ID		 3362 ceph_msg_free(m);
3363}
3364
3365struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3366{
3367 dout("%s %p (was %d)\n", __func__, msg,
3368 atomic_read(&msg->kref.refcount));
3369 kref_get(&msg->kref);
3370 return msg;
3371}
3372EXPORT_SYMBOL(ceph_msg_get);
3373
3374void ceph_msg_put(struct ceph_msg *msg)
3375{
3376 dout("%s %p (was %d)\n", __func__, msg,
3377 atomic_read(&msg->kref.refcount));
3378 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3379}
0215e44b 3380EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3381
3382void ceph_msg_dump(struct ceph_msg *msg)
3383{
3cea4c30
ID		 3384 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3385 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3386 print_hex_dump(KERN_DEBUG, "header: ",
3387 DUMP_PREFIX_OFFSET, 16, 1,
3388 &msg->hdr, sizeof(msg->hdr), true);
3389 print_hex_dump(KERN_DEBUG, " front: ",
3390 DUMP_PREFIX_OFFSET, 16, 1,
3391 msg->front.iov_base, msg->front.iov_len, true);
3392 if (msg->middle)
3393 print_hex_dump(KERN_DEBUG, "middle: ",
3394 DUMP_PREFIX_OFFSET, 16, 1,
3395 msg->middle->vec.iov_base,
3396 msg->middle->vec.iov_len, true);
3397 print_hex_dump(KERN_DEBUG, "footer: ",
3398 DUMP_PREFIX_OFFSET, 16, 1,
3399 &msg->footer, sizeof(msg->footer), true);
3400}
3d14c5d2 3401EXPORT_SYMBOL(ceph_msg_dump);
Linux kernel - Deliverables
This page took 0.435931 seconds and 5 git commands to generate.